Professional Documents
Culture Documents
Espoo 2007
Tiina M. Komulainen
Tiina M. Komulainen
Dissertation for the degree of Doctor of Science in Technology to be presented with due permission
of the Department of Chemical Technology for public examination and debate in Auditorium V1 at
Helsinki University of Technology (Espoo, Finland) on the 30th of November, 2007, at 12 noon.
Teknillinen korkeakoulu
Kemian tekniikan osasto
Prosessien ohjauksen ja automaation laboratorio
Distribution:
Helsinki University of Technology
Department of Chemical Technology
Laboratory of Process Control and Automation
P.O. Box 6100
FI - 02015 TKK
FINLAND
URL: http://kepo.tkk.fi/
Tel. +358-9-451 3852
Fax +358-9-451 3854
E-mail: tiina.komulainen@iki.fi
ISBN 978-951-22-9055-0
ISBN 978-951-22-9056-7 (PDF)
ISSN 1795-2239
ISSN 1795-4584 (PDF)
URL: http://lib.tkk.fi/Diss/2007/isbn9789512290567/
TKK-DISS-2370
Yliopistopaino
Helsinki 2007
HELSINKI UNIVERSITY OF TECHNOLOGY
AB
ABSTRACT OF DOCTORAL DISSERTATION P.O. BOX 1000, FI-02015 TKK
http://www.tkk.fi
Author Tiina Marjatta Komulainen
Department KE
Laboratory Process control and automation
Field of research Process control
Opponent(s) Prof. Maria Teresa Carvalho and Prof. Aldo Cipriano
Supervisor Prof. Sirkka-Liisa Jämsä-Jounela
Instructor
Abstract
More than 20% of the cathode copper is annually produced by copper leaching, solvent extraction and electrowinning
processes. The focus in process technology has been on research and capital intensive development of the process
equipment and chemicals. However, the financial benefits gained through an advanced control system would be significant.
An advanced control system would maximize the total production and increase the fraction of the first quality copper
cathodes by running the process closer to the optimal operating point, and by decreasing the variation in the key process
variables. The lack of adequate dynamic process models for industrial applications has, to date, prevented the development
of advanced process control systems. Therefore, the first aim of this thesis is to develop dynamic models to describe the
behaviour of an industrial copper solvent process and to facilitate control system development. The second aim of this
thesis is to develop an advanced control system for the copper solvent extraction process, and to verify that the
performance and profitability of an industrial copper solvent extraction process can be significantly increased by utilizing
the advanced process control system. In the process model, the mass transfer of copper in the mixer-settler units is
described by means of dynamic, modified ideal mixing and plug flow models. The equilibrium value for the ideal mixing
model is calculated on the basis of the steady state McCabe-Thiele diagram. The unit process models are combined
according to the case plant flowsheet and implemented into a process simulator. Based on the process models, control
hierarchy is developed. The optimization level consists of an optimization algorithm and the stabilizing control level
consists of a single input-single output control strategy employing two PI controllers or, alternatively, a multi input-multi
output control strategy using the model predictive control (MPC). The dynamic models are tested with two data sets
representing the normal operation of the industrial case copper solvent extraction plant. The models follow the output
copper concentration trends smoothly for the major input changes in the flow rates and copper concentrations, and the
residuals between the simulated data and the industrial measurements are sufficiently small, on average 1-3%. The
performance of the control system is tested in simulation environment using the industrial data as input. Compared to
manual control, the variation in the rich electrolyte copper concentration was decreased by 70 – 80 % with the PI
controllers and 80 – 90 % with the model predictive controller. The copper mass production was increased by about 3 – 5
% with both control strategies. The modeling and control results are very encouraging for the further testing of the control
system in an industrial copper solvent extraction plant.
Keywords Process Modeling, Process Control, Industrial Application, Copper Solvent Extraction, Hydrometallurgy,
Model Predictive Control (MPC), Optimization, Simulation.
ISBN (printed) 978-951-22-9055-0 ISSN (printed) 1795-2239
Print distribution Helsinki University of Technology, Laboratory of Process Control and Automation
Väitöskirjan nimi
Novel modeling and control approach for performance improvement of an industrial copper solvent extraction process
Käsikirjoituksen päivämäärä 21.5.2007 Korjatun käsikirjoituksen päivämäärä 9.10.2007
Osasto KE
Laboratorio Prosessien ohjauksen ja automaation laboratorio
Tutkimusala Prosessien ohjaus
Vastaväittäjä(t) Prof. Maria Teresa Carvalho ja Prof. Aldo Cipriano
Työn valvoja Prof. Sirkka-Liisa Jämsä-Jounela
Työn ohjaaja
Tiivistelmä
Katodikuparista vuosittain yli 20 % tuotetaan kuparin liuotus - uutto - elektrolyysi - menetelmällä. Prosessiteknologian
painopisteenä on ollut prosessilaitteiden ja -kemikaalien kehittäminen, joka vaatii paljon tutkimusta ja runsaasti
taloudellista pääomaa. Kehittyneitä säätömenetelmiä soveltamalla voitaisiin saavuttaa merkittäviä taloudellisia hyötyjä
kuluttamalla huomattavasti vähemmän resursseja. Kehittyneellä säätöjärjestelmällä kuparin uuttoprosessin tuotantoa sekä
parhaan katodilaadun osuutta koko tuotannosta voidaan kasvatettaa, koska prosessia voidaan ajaa lähempänä optimaalista
operointipistettä ja prosessin laatuparametrien vaihtelua voidaan pienentää. Kuparin uuttoprosessille soveltuvien
dynaamisten mallien puute on tähän mennessä estänyt ylemmän tason säätöjärjestemien kehityksen.
Tämän työn ensimmäinen tavoite oli kehitettää dynaaminen malli, joka kuvaa teollisen kuparin uuttoprosessin
käyttäytymistä ja mahdollistaa säätöjärjestelmän kehityksen. Työn toisena tavoitteena oli kehittää säätöjärjestelmä ja
todistaa että sen avulla kuparin uuttoprosessin tehokkuutta ja tuottavuutta voidaan merkittävästi parantaa.
Dynaaminen prosessimalli perustuu kuparin massansiirrolle vesi- ja orgaanisen faasin välilla sekoitus-selkeytys-
prosessilaitteissa. Sekoittimen malli perustuu modifioituun ideaalisekoitin malliin, jossa tasapaino lasketaan McCabe-
Thiele diagrammista. Selkeyttimen malli perustuu tulppavirtaukseen. Prosessimalli laadittiin yhdistelemällä sekoitin-
selkeytin-yksikkömalleja teollisen sovellusprosessin virtauskaavion mukaisesti. Tämän mallin perusteella implementoitiin
prosessisimulaattori. Prosessimalliin perustuen prosessille kehitettiin säätöhierarkia, joka koostuu optimointitason
optimointialgoritmista ja stabiloivan säädön tason yksikkösäätöstrategiasta (PI- ja myötäkytkentäsäätimet) tai
monimuuttujasäätöstrategiasta (MPC-säädin). Dynaamisten malleilla saatua simulointidataa verrattiin teolliseen
prosessidataan hyvin tuloksin. Simuloitu data seurasi kuparikonsentraatioiden ja virtausten vaihteluista johtuvia muutoksia
ja keskimääräinen ero simuloidun ja oikean datan valilla oli noin 1-3% testatuissa kahdessa datajoukossa.
Säätösysteemiä testattiin prosessisimulaattorilla, jonka sisääntuloina käytettiin prosessidataa. Manuaaliseen säätöön
verrattuna, rikkaan elektrolyytin kuparikonsentraation vaihtelu pienentyi 70-80% PI-säätimillä ja 80-90% MPC-säätimellä.
Kuparin tuotanto kasvoi 3-5% käyttämällä kumpaakin säätöstrategiaa. Mallitus - ja säätötulokset luovat vahvan pohjan
säätösysteemin testaukselle teollisella kuparin uuttolaitoksella
Asiasanat Prosessien mallitus, Prosessien ohjaus, Teollinen sovellus, Kuparin Uuttoprosessi, Hydrometallurgia, Malli
Prediktiivinen Säätö (MPC), Optimointi, Simulointi.
ISBN (painettu) 978-951-22-9055-0 ISSN (painettu) 1795-2239
PREFACE
The research work presented in this thesis has been carried out in the Laboratory of
Process Control and Automation, Helsinki University of Technology, between August
2003 and November 2006. I would like to thank my supervisor Sirkka-Liisa Jämsä-
Jounela for encouraging me to start PhD studies, giving me the opportunity to work
independently in an industrially related research project, and for her help with the
writing. Professor Raimo Ylinen and Dr. Mats Nikus are thanked for all their valuable
advice and discussions.
I’m very grateful to Professor Frank J. Doyle III for welcoming me into his research
group in University of California at Santa Barbara. Professor Doyle’s supervision and
collaboration with the group was really inspiring and productive. I wish to thank the
group for the great atmosphere during the unforgettable year (2005).
I would like to thank the pre-examiners of this thesis, Prof. Aldo Cipriano and Prof.
Erkki Paatero for their thorough review of the thesis and for their stimulating
comments.
The thesis work was done in collaboration with Outotec Oy. I would like to thank Ari
Rantala, Ville Suontaka and Dr. Kari Saloheimo for their support with the project and
help with the industrial aspects of the work. I wish to thank the personnel of Phelps
Dodge Morenci Inc. for great co-operation during my visits on the plant.
The thesis work was mainly funded by the Finnish Graduate School in Chemical
Engineering (GSCE). Additional funding for the research in UCSB was provided by
Jenny and Antti Wihuri foundation, Tekniikan Edistämissäätiö, Alexander and Lucie
Lampén foundation and the Foundation of Helsinki University of Technology. All
these contributions are gratefully acknowledged.
I would like to thank all my colleagues in “KEPO labra”, especially Jerri Kämpe,
Antti Remes, Ville Suontaka and Nikolai Vatanski, for keeping up the spirits through
thick and thin. I warmly thank my parents, siblings and friends for all their support
and caring over these years. Finally, my special thanks go to my fiancé Espen
Storkaas for keeping a smile on my face and encouraging me to finish this thesis.
Oslo 25.10.2007
Tiina Komulainen
2/188
ABSTRACT
More than 20% of the cathode copper is annually produced by copper leaching,
solvent extraction and electrowinning processes. The focus in process technology has
been on research and capital intensive development of the process equipment and
chemicals. However, the financial benefits gained through an advanced control
system would be significant. An advanced control system would maximize production
by running the process closer to the optimal operating point, and increase the
production of the first quality copper cathode by decreasing the variation in key
process variables. The lack of adequate dynamic process models for industrial
applications has, to date, prevented the development of advanced process control
systems.
Therefore, the first aim of this thesis is to develop dynamic models to describe the
behaviour of an industrial copper solvent process and to facilitate control system
development. The second aim of this thesis is to develop an advanced control system
for the copper solvent extraction process, and to verify that the performance and
profitability of an industrial copper solvent extraction process can be significantly
increased by utilizing the advanced process control system.
In the process model, the mass transfer of copper in the mixer-settler units is
described by means of dynamic, modified ideal mixing and plug flow models. The
equilibrium value for the ideal mixing model is calculated on the basis of the steady
state McCabe-Thiele diagram. The model utilizes industrial online and offline
measurements. The unit process models are combined according to the case plant
flowsheet. Based on the process models, a control hierarchy is developed for the case
copper solvent extraction process. The optimization level in the hierarchy consists of
an optimization algorithm that maximizes the production of the copper solvent
extraction process and provides setpoints for the stabilizing control level. The
stabilizing control level consists of a single input-single output control strategy
employing two PI controllers or, alternatively, a multi input-multi output control
strategy using the model predictive control (MPC).
The dynamic models are tested by comparing the simulated data with the industrial
data. The controller performances are tested for setpoint tracking and disturbance
rejection in the simulation environment with step input changes. The benefits of the
control system are assessed by comparing the variation in the controlled variables and
the total copper production to the data collected from the process under manual
control.
The dynamic models are tested with two data sets representing the normal operation
of the industrial case copper solvent extraction plant. The models follow the output
copper concentration trends smoothly for the major input changes in the flow rates
and copper concentrations, and the residuals between the simulated data and the
industrial measurements are sufficiently small. The average absolute error is 1-3% of
the mean value of the output copper concentrations.
The performance of the control system for setpoint tracking and disturbance rejection
is very good. As expected, the model predictive controller performs better than the PI
controllers. The disturbance rejection capabilities are further improved by adding four
feedforward compensators to the control strategies. Compared to manual control, the
variation in the rich electrolyte copper concentration was decreased by 70 – 80 % with
3/188
the PI controllers and 80 – 90 % with the model predictive controller. The copper
mass production was increased by about 3 – 5 % with both control strategies.
The modeling and control results are very encouraging for the further testing of the
control system in an industrial copper solvent extraction plant.
4/188
TABLE OF CONTENTS
PREFACE……………………………………………………………………….1
ABSTRACT……………………………………………………………………..2
LIST OF THE SYMBOLS AND ABBREVIATIONS………………………….8
1 INTRODUCTION ...............................................................................................12
1.1 Research problem and asserted hypothesis..................................................13
1.2 Scope and content of the thesis work...........................................................15
1.3 Contribution of the author............................................................................16
2 DESCRIPTION OF THE COPPER SOLVENT EXTRACTION PROCESS ....17
2.1 Hydrometallurgical processes......................................................................17
2.2 Copper leaching – solvent extraction – electrowinning process..................18
2.3 Copper solvent extraction process ...............................................................19
2.3.1 The main chemical reaction in the copper solvent extraction process.20
2.3.2 The chemical and physical side reactions in the copper solvent
extraction process.................................................................................................21
2.3.3 Process chemicals, equipment and plant configurations of the copper
solvent extraction process ....................................................................................22
3 MODELING AND CONTROL OF THE COPPER SOLVENT EXTRACTION
PROCESS – STATE OF THE ART............................................................................25
3.1 Steady state models of the mixer-settler cascades .......................................25
3.2 Dynamic models of the mixer-settler cascades............................................29
3.3 Control of the copper solvent extraction process.........................................32
4 MODEL DEVELOPMENT FOR THE COPPER SOLVENT EXTRACTION
PROCESS ....................................................................................................................34
4.1 Description of a copper solvent extraction process .....................................34
4.2 Equilibrium state model of the copper solvent extraction process ..............35
4.2.1 Estimation of the equilibrium isotherms..............................................40
4.2.2 Estimation of the extraction and stripping efficiencies........................41
4.2.3 Estimation of the recycle corrections...................................................41
4.3 Dynamic models of the copper solvent extraction process..........................42
4.3.1 Calculation of the mass transfer coefficients .......................................44
4.3.2 Estimation of the mixer and settler phase volumes .............................44
4.3.3 Estimation of the time delay for the plug flow model .........................44
5 DESCRIPTION AND THE PRELIMINARY ANALYSIS OF THE
INDUSTRIAL CASE PROCESS ................................................................................46
5.1 Description of the industrial case copper solvent extraction process ..........46
5/188
5.2 Preliminary analysis of the case copper solvent extraction process ............48
5.2.1 Data preprocessing...............................................................................49
5.2.2 Operating points of the case process....................................................51
5.2.3 Analysis of the process variation .........................................................55
5.2.4 Remarks on the quality of the process data .........................................56
6 MODELS AND THE DYNAMIC SIMULATOR FOR THE INDUSTRIAL
CASE PROCESS .........................................................................................................57
6.1 The equilibrium state models of the case process........................................57
6.2 The dynamic models of the case process.....................................................58
6.3 Dynamic simulator for the industrial case process ......................................60
6.3.1 Overall structure of the simulator ........................................................61
6.3.2 Overall structure of the unit process simulation model .......................62
6.3.3 Detailed structure of the unit process simulation model......................63
7 SIMULATION RESULTS OF THE COPPER SOLVENT EXTRACTION
MODEL .......................................................................................................................66
7.1 Input data for the simulation study ..............................................................66
7.2 Estimation of the simulation model parameters...........................................69
7.2.1 Estimation of the equilibrium isotherm parameters.............................69
7.2.2 Estimation of the efficiency and recycle correction parameters ..........72
7.2.3 Estimation of the other parameters ......................................................77
7.3 Simulation results.........................................................................................78
7.3.1 Model structures and methods for the simulation study ......................78
7.3.2 Simulation performance for the first test period ..................................79
7.3.3 Simulation performance for the second test period .............................82
7.3.4 Simulation performance under input changes......................................86
7.4 Concluding remarks .....................................................................................92
8 LINEAR MODELS OF THE COPPER SOLVENT EXTRACTION PROCESS ..
..............................................................................................................................94
8.1 Linearity of the dynamic models .................................................................94
8.1.1 Scaling invariance................................................................................94
8.1.2 Additivity ...........................................................................................101
8.1.3 Summary of the linearity study..........................................................105
8.2 Determining the order of the linearized process models ...........................106
8.2.1 Linear model of one unit process.......................................................106
8.2.2 Linear model of the copper solvent extraction process .....................107
8.3 Identification of the linear models .............................................................110
6/188
aq aqueous
el electrolyte
entr entrainment
in into a mixer
int interface between aqueous and organic solutions
max maximum value
min minimum value
o, op operating point
opt optimum value
org organic
out out of mixer / unit
outs out of settler
ABBREVIATIONS
10/188
1 INTRODUCTION
Copper (Cu) and its alloys are widely utilized in a great variety of industrial
machinery and equipment, infrastructure tubes and cables, and a range of consumer
products such as cars and cell phones. Copper mainly occurs in copper-iron sulphide
and copper sulphide mineral form. The largest producers of copper ore are Chile, the
USA, Indonesia, Peru, Australia, Russia and China. (Moskalyk and Alfantazi, 2003,
Norton and Leahy, 2006)
The world production of copper in 2004 was about 15.800 metric tons, of which 88 %
was from primary copper refining and 12% from the treatment of copper-containing
scrap. The main refining options are hydrometallurgical and pyrometallurgical
processes. Currently, around 20 % of the copper ores are processed by
hydrometallurgical methods, with copper cathode production accounting for over
2.700 metric tons annually. The largest producers of copper by leaching, solvent
extraction and electrowinning processes are Chile (1.636 metric tons), the USA (584
metric tons), Peru (167 metric tons) and Zambia (83 metric tons) (Edelstein 2004)
Hydrometallurgical treatment is preferred for oxidized copper ores such as cuprite
(Cu2O) and low grade secondary ores such as chalcocite (Cu2S). In the future, the
most common copper ore, chalcopyrite (CuFeS2), can also be industrially processed
by hydrometallurgical methods with improved leaching techniques, such as pressure
leaching and bio-oxidation leaching. The oxidized copper ores are found in South
America, the southwest of North America, Australia and Southern Africa. A typical
hydrometallurgical processing chain for copper ore consists of leaching, solvent
extraction and electrowinning. Copper rich slag from other metallurgical processes
can also be treated by hydrometallurgical methods. (Habashi, 1999, Seward, 1999,
Kordosky, 2002, Munnik et al., 2003, Rydberg et al., 2004, Hyvärinen and
Hämäläinen, 2005).
The importance of hydrometallurgical copper production is increasing strongly:
between 2000 and 2004 hydrometallurgical copper production increased in Chile by
19%, in the USA by 5%, in Peru by 31% and in Zambia by 27%. Since the industrial
production of copper by the leaching-solvent extraction and the electrowinning
process started in the USA in around 1970, the proportion of hydrometallurgical
copper production out of the total mined copper production has increased from 1,2 %
to 38% in 30 years. The reason for this is the depletion of the rich, copper ore bodies
that provide suitable raw material for pyrometallurgical processing, and the tightening
of environmental requirements which favour hydrometallurgical processing over
pyrometallurgical processing. (Bartos, 2002, Kordosky, 2002, Edelstein 2004)
Therefore, the development of new technologies to increase the performance and
profitability of industrial copper leaching, solvent extraction and electrowinning
processes is very important. During the past decade, the development has focused has
on process technology. The leaching efficiency has been increased by the new
leaching processes, for example pressure leaching (Hyvärinen and Hämäläinen, 2005)
and bioleaching. New chemicals and equipment have also enhanced the performance
of the solvent extraction and electrowinning processes. (Habashi, 1999, Kordosky,
2002).
However, the major challenge in the field of copper leaching-solvent extraction-
electrowinning, the development of an advanced control system for industrial plants,
13/188
has still not been resolved (Bergh et al., 2001, Bergh and Yianatos, 2001, Jämsä-
Jounela, 2001, Hodouin et. al., 2001). Basic control level PID controllers and basic
measurements, possibly with online copper and impurity concentration analyzers, are
currently used in most plants (Flintoff, 1993, Hughes, D., Saloheimo, K., 2000, Bergh
and Yianatos, 2001). However, the lack of suitable dynamic models for industrial
applications has prevented the development of advanced process control systems for
the copper solvent extraction process. A considerable amount of work was carried out
on the modeling laboratory-scale solvent extraction mixer-settler cascades in the
1970s and 1980s but, due to computational restrictions and inadequate industrial
online measurements, model development did not reach the stage of industrial
applications (Wilkinson and Ingham, 1983, Ingham et al., 1994). Recently, Aminian
et al. (2000) published a steady-state model for a copper solvent extraction and
electrowinning pilot plant, and Komulainen et al. (2005, 2006) a dynamic model for
an industrial copper solvent extraction process.
The primary control aims are the maximization of production, rejection of
disturbances in the leach solution copper concentration and pH level, a reduction of
process variation, and fast adaptation to the most optimal operating point. An
advanced control system with higher level optimizer should be developed in order to
meet these aims.
Due to the long time delays and complex interactions in the copper solvent extraction-
electrowinning process, one possible multi-input multi-output (MIMO) approach
would be model predictive control (MPC). MPC is effective in handling time delays
and constraints, and the future process outputs, calculated on the basis of the process
model, can be easily visualized for the process operators. MPC is widely used in
several application areas in the process industries, including the refining, chemical,
and papermaking industries (Qin and Badgwell, 2003). MPC applications have
recently emerged in the metal and mineral industries, for example for the control of a
flotation bank (Hodouin et al., 2000) and the control of a pilot flotation column
(Nunez et. al. 2006).
The economic benefits of maximizing the total copper production and minimizing the
production costs by process control methods would be significant for industrial
copper solvent extraction plants. Assuming that the control system does not increase
production costs and the copper price is 1,67$/lbs (average LME 2005), a production
increase of 5% would be equivalent to 60 million dollars per year for a copper solvent
extraction-electrowinning plant with an annual production of 365000 tonnes.
disturbances from the leach solution. Thus, an advanced process control system would
maximize total production, increase the production of the best (grade A) quality
copper cathodes, while minimizing the consumption of chemicals and energy per
produced copper tonne.
The focus in this thesis is on performance improvement of the copper solvent
extraction process through the utilization of process control methods. In order to
investigate the benefits of a process control system for an industrial copper solvent
extraction process, this thesis aims at developing dynamic process models and, based
on these, an advanced control system for the copper solvent extraction process.
The hypotheses of the thesis are:
1. Dynamic process models utilizing only industrial online process measurements are
able to describe the behaviour of an industrial copper solvent extraction process. The
dynamic models enable development of an advanced process control system for the
copper solvent extraction process.
2. Utilization of the advanced process control strategy significantly increases the
performance and profitability of an industrial copper solvent extraction process
compared to the utilization of a manual control strategy.
In order to prove the hypotheses, the following tasks have to be performed:
1. To develop static and dynamic models of the copper solvent extraction
process.
2. To construct a process simulator on the basis of the process models, and to
verify the modeling accuracy with industrial data.
3. To linearize the dynamic models, and to design the control strategy on the
basis of these linear models.
4. To develop an optimization algorithm for maximizing the copper production
5. To design single input-single output (SISO) controllers for set-point tracking
and disturbance rejection.
6. To design a multi input-multi output (MIMO) controller, and to compare the
SISO and MIMO control strategies on set-point tracking and disturbance rejection.
7. To test the controller performance with industrial data. In order to assess the
benefits of the control system, the controller performances in the simulation
environment are compared to the data from an industrial plant under manual control.
The first and second tasks increase our understanding of the dynamic process
phenomena, enable control system development, and provide a test bench for the
control strategies. The third to sixth tasks provide the optimizer and the SISO and
MIMO control strategies for the industrial copper solvent extraction process. The
profitability and performance of the control system are asserted within the seventh
task.
15/188
ORE
Size reduction
Concentration
Activation
Leaching
Solution purification
Precipitation
PRODUCT
Figure 2-1: General flowsheet of hydrometallurgical processing [Aromaa, 1990, Hayes, 1993].
The mined ore, or other feed material, is first pretreated by crushing, concentrating
(e.g. flotation), activation by chemical changes and physical modifications (e.g.
agglomeration) or roasting. Next, the process options in the leaching stage are
oxidative leaching, acid leaching, alkaline leaching, metathetic leaching, irrigation
leaching (in situ, dump and heap), percolation leaching, and bacterial leaching. If the
liquid contains solid particles after leaching, separation can be done by filtration,
thickening and clarification (including coagulation, flocculation), counter-current
decantation or with other equipment like centrifuges and hydrocyclones. Solution
purification techniques include precipitation of impurities (pH controlled,
hydrothermal, crystallization, cementation and electrolytic processes), solvent
extraction, adsorption and ion exchange, membrane processes (reverse osmosis,
ultrafiltration and electrodialysis), and chemical reactions (oxyhydrolysis). Process
options for metal recovery are electrowinning, chemical reduction, precipitation and
crystallization. The hydrometallurgical processing chain of the different unit
processes and conditions are designed on the basis of the type of raw material. Each
type of ore, slag matte and recycled material requires a tailored processing option.
(Ray et al., 1985, Aromaa, 1990, Hayes 1993, Mooiman et al. 2005)
The copper concentration varies from less than 1 g/l to 35 g/l, and the pH varies
between 0,8 and 2,5. The main impurities are iron, manganese, chloride and nitrate.
The second phase in the hydrometallurgical treatment of copper ore is solvent
extraction. In copper solvent extraction, the aqueous copper leach solution is
concentrated and purified using a copper-selective organic solution. The solvent
extraction process consists of extraction and stripping processes, both of which may
contain parallel and series unit processes. Currently, all the industrial operations use
mixer-settler type of equipment, in which the solutions flow in a counter current
direction. In mixer units, the phase with the smaller volume is dispersed into the
continuous phase in order to ensure a maximal mass transfer interface. The mixer
units consist of pumping and mixing parts, with axial impellers designed to maintain
the dispersion. The aqueous and organic phases are separated in settlers. Settlers are
long and shallow in oder to ensure proper phase separation, thus minimizing
entrainment. The organic solution is a mix of an organic solvent such as kerosene and
the copper selective extractant. Industrially used extractants are ketoximes and
aldoximes, possibly with modifiers, and mixtures of these. The extractant blend is
chosen according to the leach solution, operating conditions and plant design.
(Kordosky, 2002, Robinson, 2003, Vancas, 2003, Rydberg et al., 2004)
The third phase in the hydrometallurgical treatment of copper ore is electrowinning.
In this process, the concentrated aqueous copper solution, the rich electrolyte, is
electrowon to form 99,999% pure copper cathodes. In electrowinning, the copper
concentration of the electrolyte is typically between 32 and 37 g/l, the sulphuric acid
concentration between 160 to 180 g/l, and the current density around 240 – 320 A/m2.
The current efficiency can be up to 95%. Most electrowinning operations use rolled
anodes of Pb-Ca or Pb-Sr-Sn due to their low corrosion rate and dimensional stability.
Stainless steel blanks are used as starter cathodes to improve current efficiency and
copper quality. In the larger plants, the copper cathodes are mechanically stripped
from the steel blanks. (Bergh and Yianatos, 2001, Kordosky, 2002)
Examples of the copper solvent extraction and elecrowinning processes can be found
in Sole et al. (2005), Munnik et al. (2003) and Whyte et al. (2001), Jenkins et al.
(1999).
Leach Electrolyte
solution solution
Organic
solution
Electro
Leaching Extraction Stripping Cu
winning
The main chemical reaction taking place in the copper solvent extraction process is
described in Section 2.3.1, and the chemical and physical side reactions of the copper
solvent extraction process in Section 2.3.2. The process chemicals, equipment and
plant configurations used in industrial copper solvent extraction plants are discussed
in Section 2.3.3.
2.3.1 The main chemical reaction in the copper solvent extraction
process
In the copper solvent extraction process, the main phenomenon is the mass transfer of
copper between the aqueous and organic solutions. The mass transfer of copper is
based on cation exchange reaction, where the copper ions are forming chelates with
copper selective reagent molecules. The equilibrium of this cation exchange reaction
can be affected by adjusting the acidity of the aqueous solution.
The equilibrium reaction is as follows:
Cu 2+ (aq ) + SO42− (aq ) + 2 RH (org ) ↔ CuR2 (org ) + 2 H + (aq ) + SO42− (aq ) (2-1)
In the extraction units, the copper ions form complexes with reagent molecules in an
acid (pH ~1.8, ~5 g/l H2SO4 in PLS solution (Kordosky, 2002) over the temperature
range 15 – 25 °C (Kordosky, 2006). The copper chelate is very stable in as low pH as
1.8. The proper pH range for extraction depends on the reagent characteristics, ferric
concentration and the other impurities in the leach solution. Each reagent has an
optimal pH for the most selective extraction of copper from the other metals present
in the leach solution. Too low a pH can change the equilibrium reaction to prefer
breaking of the chelate.
The copper-rich organic solution is led from extraction to the stripping units, where
the strongly acid environment (~160 – 180 g/l H2SO4, Kordosky, 2002) catalyses
disruption of most of the organic copper complexes. The chelate breaking reaction
can be presented as follows:
CuR2 (org ) + 2 H + (aq ) + SO42 − (aq ) → Cu 2 + (aq ) + SO42− (aq ) + 2 RH (org ) (2-2)
In the stripping units, high acidity is preferred in order to achieve fast mass transfer.
However, due to the increased amount of acid mist in the downstream electrowinning
facilities and the corrosive effect on the equipment in the electrowinning process, the
acidity should remain under 200 g/l.
21/188
In the copper solvent extraction process it is essential to maintain the optimum within
the copper mass balance; the amount of copper extracted from the leach solution in
the extraction units has to be stripped into the electrolyte solution in the stripping
units.
The exact equilibrium for the liquid-liquid extraction system can be calculated from
the ternary phase data or equilateral-triangular diagrams (Robbins, 1984). Assuming
the solubility of organic and aqueous phases can be neglected, simpler methods can be
utilized. Thus, in industrial practice methods, such as the McCabeThiele diagram, also
known as the Fenske diagram, are widely used (Mills, 1983, Robbins, 1984, Hayes,
1993).
The equilibrium copper solvent extraction unit processes can be studied by
formulating an equilibrium constant κ for the system:
[CuR2 ] ⋅ H +
2
κ= (2-3)
Cu 2 + ⋅ [ RH ]
2
where [CuR2] is the copper complex concentration in the organic phase, [H+] the
hydrogen concentration in the aqueous phase, [Cu2+] the copper concentration in the
aqueous phase and [RH] the reagent concentration in the organic phase. (Hayes, 1994)
The kinetics of the solvent extraction includes both the mass transfer and the chemical
reaction in the heterogeneous system. The extraction is controlled by the diffusion or
chemical reaction rate. Diffusion is affected by the mass transfer surface area and
concentration of the slowly diffunded reagents. The rate of the chemical reactions on
the phase surface is affected by the surface area, the activity of the reacting
components, and the molecular geometry.
In the copper solvent extraction process, the chemical reaction rate is affected by the
solubility of both phases, the distribution and ionization coefficients, and the volumes
of both phases. The reaction rate is defined by:
x y
k f Cu 2+ R −
r= z
(2-4)
H +
, where kf is the rate constant for the forward reaction, [H+] the hydrogen
concentration in the aqueous phase, [Cu2+] the copper concentration in the aqueous
phase and [R-] the reagent concentration in the organic phase.
2.3.2 The chemical and physical side reactions in the copper
solvent extraction process
The main chemical side reaction is the extraction of iron from the leach solution into
the organic solution in the extraction units. Only ferric (Fe3+) iron forms complexes
with the organic solution, and it can therefore be transferred chemically from the
leach solution via the organic solution to the electrolyte solution. The reaction rate is
dependent on the acidity of the leach solution and the selectivity of the organic
solution. The ferrous and ferric ions accumulate in the electrolyte and prevent
maximum possible copper transfer from the organic solution to the electrolyte
22/188
solution. To prevent ferrous and ferric loading of the electrolyte, small amounts of the
electrolyte are bled to the extraction units.
The main physical side phenomena are the transfer of impurities with entrainment and
the accumulation of crud. Ferrous (Fe2+), manganese (Mn2+), and other impurities can
pass from the leach solution to the electrolyte solution only as aqueous entrainment,
i.e. in aqueous droplets within the imperfectly separated organic solution from the
extraction units to the stripping units. In electrolysis, the presence of iron reduces
current efficiency and decreases copper production. Manganese is converted to
permanganate, which circulates back to stripping, and disrupts the reagent in the
organic phase. Since the reagent is the most expensive chemical in the plant, this
causes severe financial losses. In order to minimize such losses, a mixer-settler unit,
called the washing or scrubbing stage, is added between the extraction and stripping
processes. The aim of the washing stage is to remove the entrained PLS solution
containing impurities and to strip the ferric ions from the organic solution. A
concurrent strategy is to bleed small portions of the electrolyte solution from the
electrowinning process to the washing stage or one of the extraction stages in order to
remove the excess iron. The choice of reagent, copper and acid concentration in the
wash stage aqueous solution, and the mixer retention time, also affect the transfer of
impurities in the solvent extraction process. (Kordosky et. al., 2000, Virnig and
Olafson, 2002)
All the solvent extraction plants contain at least small amounts of crud, gelatinous or
crystalline emulsions of solid particles, in the aqueous and organic solutions. The
solid particles and biological material in the leach solution, as well as precipitates and
silica generated during changes in the pH in the mixers, are sources of crud formation.
Crud formation is especially fast in a low pH range. A small amount of crud improves
separation of the dispersion of the organic and aqueous phases in the settler, but
excess crud causes severe problems. (Biswas and Davenport, 1994, Kordosky, 2002,
Ritcey, 2002, Tetlow 2003)
2.3.3 Process chemicals, equipment and plant configurations of
the copper solvent extraction process
The organic solution in the copper solvent extraction process consists of a solvent and
a reagent. The high viscosity of the reagents means that they have to be diluted in the
organic solvents. The preferred solvent characteristics are a flash point of over 60 °C,
difficultly vaporizable, and a specific gravity of around 0,8 in order to ensure good
separation from the aqueous solution. The solvents are usually different mixtures of
paraffin, and aromatic and naphthalene hydrocarbons like kerosene. The aromatic
compounds increase copper solubility to the organic solution, but slow the kinetics of
extraction and stripping. (Biswas and Davenport 1994, Kordosky, 2002)
The reagents are typically hydroxy oximes: salicyl aldoximes, ketoximes and their
mixtures. Ketoximes are readily soluble in kerosene, resistant to heating and do not
require modifiers. Compared to the aldoximes, a lower acidity can be used in
stripping ,and small amounts of particles, like colloid silicates and flocculents, are
allowed in the pregnant leach solution. However, the extraction strength, kinetics and
selectivity for ferric are inferior to those with aldoximes. The salicyl aldoximes are
strong extraction chemicals with good extraction kinetics and selectivity against
ferric. The aldoximes dissolve easily in kerosene, and separate rapidly from the acid
aqueous solutions. Stripping requires such a high acidity that equilibrium modifiers,
such as tridecanol and nonylphenol, are necessary. Modifiers prevent decomposition
23/188
of the aldoximes and form stable complexes with silicates possibly present in the
leach solution. Possible side effects are crud formation, organic losses due to
entrainment in the raffinate leach solution, and aqueous entrainment to stripping.
Mixtures of ketoximes and aldoximes combine the strengths of these hydroxyl
oximes, good extraction capability, fast kinetics, selectivity, stability and an even
better stripping ability. In mixtures ketoximes replace aldoxime modifiers.
Commercial ketoximes include LIX64N, LIX65N, LIX84, and aldoximes LIX860,
LIX612N-LV, acorga P5100, acorga PT5050 and the mixture LIX 984. (Kordosky et
al., 1987, Biswas and Davenport, 1994)
Typical process instrumentation in industrial copper solvent extraction plants includes
the measurement of flow rate, temperature, and level. Conductivity measurements in
mixers, pH-meters, and online measurements of the copper and impurity
concentrations may exist. The copper and impurity assays, phase ratio, breaking time,
and other diagnostic measurements are performed by the process operators and the
laboratory. (Bergh and Yianatos, 2001)
The only equipment in industrial copper solvent extraction plants is the mixer-settler.
Copper transfer between the aqueous and organic phases takes place in mixers, and
the phases are separated in settlers. Different mixer designs are utilized, for example,
in the Outokumpu VSF technology the organic and aqueous flows are combined in a
dispersion pumping unit followed by two mixers (Nyman et. al. 2003). Most of the
modern settlers are large and shallow. (Kordosky 2002, Robinson, 2003)
The unit process configurations of the industrial plants vary from a simple 2 series
extraction steps with one stripping step to more complex configurations with 1
parallel extraction step, 2 series extraction steps, 1 washing step and 2 stripping steps
(Kordosky, 2002, Sole et al., 2005). The relationship between the extraction and
stripping units and the copper concentrations can be visualized using the McCabe-
Thiele diagram. (Mills, 1983, Robbins, 1984, Hayes, 1993 (p. 277))
Different design aspects of solvent extraction operation using mixer-settlers have been
studied by Galvez et al. (2004) and Pinto et al. (2004). Galvez et al. (2004) developed
a graphical method to study the different flow configurations between the unit
processes, including the organic and aqueous recycles inside the mixer-settler pairs
and bypass flows. In their approach, the slopes of the operating lines in extraction and
stripping are first transformed so as to be the same and are therefore presentable at the
same scale (e.g. 0 – 8 g/l Cu), and the analysis is then performed with the
experimentally determined equilibrium curves. The method enables easy design and
comparison of the different flowsheets.
Pinto et al. (2004) developed a complex steady state simulation model to test the
selectivity of the similarly extracted metal species under different mixing conditions.
The studied parameters included the mean residence time, the dispersion phase hold
up and the agitation speed. The optimal parameters were found using multiobjective
optimization, illustrated with a case study of zinc and cadmium extraction in a
sulphuric acid environment. This method helps in optimizing the process conditions,
but requires detailed models of the process and industrial measurements of the phase
hold up and residence time.
Cognis has developed a statistical program for designing copper solvent extraction
plants with different reagents and varying temperatures and copper and acid
concentrations. The accuracy of the statistical model was tested by comparing
24/188
measured and predicted stripped organic values obtained with the reagents LIX 612N-
LV, LIX 860N-I and LIX 8180, as described by Kordosky et al. (2006). The
agreement between the predictions and measurements under varying copper
concentrations (30-55 g/l), acid concentrations (130-180g/l) and temperatures (35 -
41°C) was very good. The copper recovery was also analyzed using the Cognis
Isocalc model for the extraction circuit. An increase in temperature increased copper
production under the assumption that the extraction efficiency is higher at higher
temperatures.
25/188
corg
Extraction equilibrium
isotherm
. .
Stripping
operating
Extraction line Stripping
equilibrium
operating line
isotherm
. .
caq
Figure 3-1: The McCabe-Thiele diagram for a process with one unit process for
both extraction and stripping, marked with the dashed triangles. The extraction
and stripping equilibrium isotherms and operating lines are marked with solid
lines. The horizontal axis is the copper concentration in the aqueous phase and
the vertical axis the copper concentration in the organic phase.
26/188
Ac aq
c org = (3-1)
c aq + B
The equilibrium isotherm for stripping can be approximated with a linear function
between the copper concentrations of the organic and aqueous solutions with
constants C and D as follows:
c org = C ⋅ c aq + D (3-2)
estimate the building and operational costs and the equipment dimensions of the plant.
The heap leaching model is a linear function based on the amount and grade of the
copper rich ore, leaching time and leaching yield. The solvent extraction model is
based on the McCabe Thiele diagram. The inputs are the PLS copper concentration
and flow rate, electrolyte copper concentration, organic to aqueous ratios in the
mixers, and mixer efficiencies. The electrowinning model is based on the cathode
area, current efficiency and growing period. The model was developed in the
Microsoft Excel environment and illustrated with the flow sheet calculation and the
cost estimate of the industrial copper leaching-solvent extraction –electrowinning
plant in Zaldivar, Chile.
A multivariable model of the electrowinning process of the Zaldivar copper leaching,
solvent extraction and electrowinning plant was developed by Katajainen (1998). The
lean electrolyte copper concentration was modeled with a linear combination of the
lagged variables of rich electrolyte copper concentration and flow rate, cathode
currents, acidity of the rich and lean electrolyte solutions, chloride concentration and
temperature of the electrolyte solution. The model fit to the operational data was good
within similar operating conditions as in the modeling data set, with a standard
deviation of 0,23. With the other data sets the standard deviation increased to 0,66.
The model gives a good starting point for dynamical model development for the
electrowinning process.
A steady state model of an industrial copper solvent extraction and electrowinning
plant was developed by Aminian et al. (2000). The model for the solvent extraction
considers mass transfer in the mixers. The inputs of the model are the copper, iron and
acid concentrations, and the aqueous and organic flow rates. The mixer model is
based on mass conservation and it includes both transfer to the interfacial surface of
the phases and the reaction rate over the surface. The equilibrium curves are
experimentally defined. The output copper concentration of the mixer unit is modeled
as:
RH n + 2− p
6V org [ int ] m
H cint
org
=c + κ −
int
corg org
k r
c aq
d B F org H + p
(3-3)
[ RH int ]
out in int 2− n aq 1−m
cint
int
where Vorg is the mixer volumefor the organic solution, Forg the organic flow rate, dB
the mean diameter of the organic droplets, kr the reaction rate constant, κ the
equilibrium constant, RH the reagent concentration and H+ the hydrogen ion
concentration at the interface. The indices n,p, and m are partial reaction orders.
The input variables of the electrowinning model are the copper, iron and acid
concentrations, electrolyte flow rates, temperatures and voltages. The steady state
model predicting the output copper concentration of the electrowinning has the
following form:
Finel el 1 M
c el
out = el cin − el Cu η I (3-4)
Fout Fout zF
where concentrations are marked with c, electrolyte flow rates with Fel, z (=2) the
charge of copper ion, F the Faraday constant, MCu the atomic weight of copper, η the
current efficiency, and I the total electric current.
28/188
The models were combined to agree with the flow sheet of the pilot plant consisting
of two series extraction units, one stripping unit and an electrowinning process. The
model parameters were first calibrated with one data set and the model was then
tested with other data set. The measurements and steady state model predictions were
compared and had good agreement. The simulator can be used to increase
understanding of the steady state process and to perform the optimization of copper
solvent extraction and electrowinning plants. The steady state models also provide a
good starting point for the development of the dynamic process models necessary for
the control studies.
A mass balance monitoring prototype with data reconciliation for an industrial copper
solvent extraction and electrowinning process has been reported by Suontaka et al.
(2004, 2003). The inputs to the monitoring model were the flow rate and copper
concentration measurements. In order to filter the measurement noise, a static data
reconciliation based on the mass conservation equations was used. The mass balance
model assumed steady state operation and constant equilibrium isotherms. The
predictions of the mass balance model were compared to the output copper
concentrations in order to monitor performance of the process. The prototype was
tested with plant history data and the operational principle of the application was
found to be successful. The aim of the monitoring system was to assist operators with
process control, and was implemented in the Microsoft Excel environment. The
system can be used to monitor other copper solvent extraction and electrowinning
plants with only small plant-specific modifications.
Since relatively few articles have been published on steady state modeling of the
copper solvent extraction process, three papers describing approaches to modeling
solvent extraction with mixer-settler equipment are referred to here. Abdeltawab et al.
(2002) studied the effect of agitation speed and aqueous to organic flow ratio on the
extraction of rare earth metals in a laboratory scale mixer-settler column (with 5
extractions, no stripping stages). The basic formulation of the steady state output
concentrations sums the input concentration with a stage efficiency weighted
difference between the equilibrium and input concentrations as follows:
[ ] c
R int
3
( )= c K aV κ (3-5)
=c +α c * −c + org o
[ ] −c
org org org org org aq org
c e + int
F + K o aV
out in out in in in in
H
The equilibrium curve between the aqueous and organic concentrations is linear,
where the coefficient consists of the equilibrium constant κe times the third order term
of the ratio between the reagent [Rint] and the hydrogen ion [H+int] concentrations at
the droplet interface. The interface concentrations lead to detailed models of the mass
transfer coefficients and the interfacial areas. The stage efficiency α is defined with an
overall mass transfer coefficient Ko, organic flow rate Forg, interfacial area, a, and
mixer volume V. The model was very successful in predicting the steady state effects
of agitation speed, aqueous feed solution pH, and the number of stages.
Nishihama et. al. (2003) developed a steady-state model for the extraction of rare
earth ions with EHPNA (2-ethylhexyl phosphonic acid mono-2-ethylhexyl ester) in
the presence of EDTA. The model, based on material balance equations and the
experimentally verified equilibrium curves, at different pHs and reagent loadings, was
developed to design a separation process with a counter-current mixer-settler cascade.
The emphasis in this steady state model was more on the process equipment design
29/188
than on further dynamic model development. Testing clearly implied that there was an
increase in extraction with an added scrubbing stage, and a mixer-settler cascade of 10
extraction steps and 1 washing (scrubbing) step was proposed for this system. As was
the case for the steady state model described by Takahashi et al. (2002), the lack of
stripping steps for the process partly prevents comparison with the industrial copper
solvent extraction process, in which the total performance is strongly connected to the
effects of organic solution recycling between the extraction and stripping steps.
υ
org org
dcout
dt
(
= F org cinorg (t ) − cout
org
) (
(t ) + κ a cout
aq
*(t ) − cinaq (t ) ) (3-6)
where υ is the stage specific organic solution hold-up in the mixer, κ the stage specific
mass transfer coefficient and a is the interfacial area of the extraction stage. In
equilibrium the driving force (cout aq
* (t ) − cinaq (t )) was recommended to be close to zero,
which implies that the term will have an arbitrary high value. The equilibrium
equation used to calculate c* was assumed to be linear and the parameters constant.
org
The settler output value cout , settler was a time-delayed value of the mixer output value,
defined as follows:
where the time delay (t- tn) is calculated by dividing the phase volume in the settler
with the flow rate (see equation 4-39).
A more exact approach was to model the hydrodynamic effect between the mixer and
settler, and to model the settler with a series of ideal mixers. In the detailed settler
model the settling volume was divided into perfectly mixed stages with back mixing
streams, separately for both phases.
In Ingham et al. (1994), the mixing model was further modified by adding an
entrainment flow, centr, which is the total flow of the organic droplets in the incoming
aqueous flow, as presented in Equation (3-8). The hold-up volume υ was assumed to
be proportional to the flow rate of the phase, and the hold-up was replaced by the
mixing volume, Vm. The equation for mixing was presented as follows:
30/188
org
( ) ( )
dcout
V org
m = F org cinorg (t ) + g m centr
org
(t ) − (1 + g m )cout
org
(t ) + κ a cout
aq
* (t ) − cinaq (t ) Vmaq
dt
(3-8)
where Vorg and Vaq are the organic and aqueous volumes in the mixer, and gm the
percentage of the entrained flow. The mass transfer coefficient κ was derived from
two film theory using the phase specific film coefficients korg and kaq, and the
derivative of the equilibrium curve corg* = f(caq*).
k aq k org
κ aq
= (3-9)
k org + k aq ⋅ f '(c aq *)
Instead of the mass transfer coefficient term κa, the Murphree stage efficiency can be
used. However, as indicated by Wilkinson and Ingham (1983), the Murphree
efficiency coefficient changes considerably more during operating point changes than
the coefficients in the previous approach. The Murphree efficiency was determined
as:
cinaq − cout
aq
E= (3-10)
cinaq − cout
aq*
The settler model consisted of plug flow and mixed flow regions, where the latter
represents the turbulent flow region and the backmixing phenomena in the settler. The
output organic concentration from the plug flow model was a time-delayed value of
the mixer output organic concentration, as in Equation (3-5). The output concentration
for the well-mixed region was calculated as follows:
org
( )
dcout
= g s ⋅ F org ⋅ cout , s (t ) − cout , m (t )
,s
Vsorg org org
(3-11)
dt
aq
( )
dcout kS
V = F aq cinaq (t ) − cout
aq
(t ) − r aq ⋅
dt F
κ cout
aq
(
(t ) c( RH )in (t ) − 2 F aq F org ⋅ cinaq (t ) − cout
aq
((t ) ))
(
c( H + )in (t ) + 2 cinaq (t ) − cout
aq
(t ) ) (3-12)
⋅
− in ( in (
out in )) ( in out (
c org (t ) + F aq F org ⋅ c aq (t ) − c aq (t ) ⋅ c( H + ) (t ) + 2 c aq (t ) − c aq (t )
))
(
c( RH )in (t ) − 2 F aq F org ⋅ cinaq (t ) − cout
aq
(t ) ( ))
where kr is the reaction rate constant, S the interfacial area, κ is the equilibrium
constant (=4.0), c(RH) the organic reagent concentration and c(H+) the hydrogen ion
31/188
concentration. If one of the phases is recycled from the settler back to the mixer, then
the krS term becomes dependent on the aqueous phase hold up ν with a polynomial
correlation, as follows:
kr S = 3.78 − 16.6ν + 28.1ν 2 − 16.2ν 3 , v ∈ [1/ 3, 0.8] (3-13)
where the aqueous hold up with the recycle flow rates, Frecycle, is determined as
follows:
Finaq + Frecycle
aq
v`= (3-14)
Finaq + Frecycle
aq
+ Finorg + Frecycle
org
The output concentrations of the mixer-settler unit are lagged with a time delay t
calculated by dividing the total volume by the sum of the flow rates.
The model was configured for an experimental system with one extraction step, one
washing step and five stripping steps. The input pregnant leach solution contained 1.6
g/l copper and 0.016M H2SO4, the organic solution contained 12.5% LIX64N reagent
in kerosene, and the aqueous stripping solution 150 g/l H2SO4. The model fit to the
experimental data was very good. On the basis of the experiments the authors suggest
that the organic to aqueous ratio be kept near 2.0, by adjusting the flow rates or the
recycle flow rates.
An example of applying Murphree efficiency to modeling solvent extraction with
mixer-settler equipment is described by Salem and Sheirah (1990). They applied a
dynamic model to describe the dynamic behaviour of the mixer-settler cascade. In the
model, a linear equilibrium isotherm dependent on the temperature and aqueous
equilibrium concentration was assumed. The experimental laboratory system
consisted of a mixer-settler cascade with four extraction steps and a water-toluene-
acetic acid medium. With the model, the effects of agitation speed, temperature and
holdups and the optimal number of extraction stages were studied. The dynamic
model gave adequate fit to the experimental data.
Aminian et al. (1998) used the previously presented dynamic mixer-settler model to
study the residence times of an extraction cascade pilot plant with two extraction and
one stripping stages. The tracer experiment for data collection was carried out with
lithium chloride. The best model structure to describe the residence time distribution
of the lithium chloride impulse response was a mixer model with two ideal mixers,
and a settler model with two parallel models, minor flow to the ideal mixing series,
and most of the flow to the plug flow in series with two ideal mixers. The
electrowinning model consisted of one perfect mixer with a very small bypass of plug
flow and two perfect mixers in series. The simulation results were very consistent
with the experimental data.
A discrete, pulsed flow model for a rare-earth solvent extraction cascade was
developed by Wichterlowa and Rod (1999). The model for one mixer-settler pair was
separated into one mixing and several settling steps, in which each of the steps had
equal time intervals. In the mixing step, the metal was transferred between the
aqueous and organic phases according to the linear equilibrium relationship. The
settling steps were modeled as ideal mixers, separately for each phase. The discrete
output concentration in the ideal mixing step with mass transfer was presented as
follows:
32/188
αˆ ⋅ m
org
cout (t + 1) = (1 − αˆ )cint
org
+ αˆ ⋅ cint
org *
= (1 − αˆ )cint
org
+
V aq
+ mV org
(V aq aq
cint + V org cint
org
)
v org c out
org
(t ) + F org ∆t ⋅ cinorg (t )
= (1 − αˆ ) (3-15)
V org
αˆ ⋅ m
+ aq
V + mV org
(
v aq cout
aq
(t ) + F aq ∆t ⋅ cinaq (t ) + v org cout
org
(t ) + F org ∆t ⋅ cinorg (t ) )
where F∆t is the pulsed volume entering each ideal mixing step, Vaq and Vorg the ideal
mixer volumes for the aqueous and organic phases with half of the entering pulsed
phase volumes, and νaq and νorg the ideal mixer phase volumes minus half of the
entering pulsed phase volume. m is the linear slope of the equilibrium curve, and α̂
the efficiency. The equilibrium curve is a nonlinear function of the steady state
efficiency, concentrations and volumes. The model was applied to neodymium and
praseodymium extraction with DEHPA in a mixer-settler cascade of 6 extraction and
6 stripping stages. The system was tested on laboratory scale extraction equipment
with 30 mixer-settlers, each having a volume of 2,5 liters. The model could be used
for the industrial copper solvent extraction process, and was characterized by
relatively stable mixer and settler volumes, flow rates and flow ratios. However,
division of the mixers and settlers into ideal mixing steps should be done carefully.
The control system for the solvent extraction and electrowinning plant has several
tasks, such as keeping the operation within the safety limits, maximizing throughput
and rejecting disturbances. These tasks, with the exception of safety-related issues,
can be formulated into a set of control objectives. Bergh and Yianatos (2001)
proposed that the main control objectives for copper solvent extraction and
electrowinning processes are maximization of the production rate while fulfilling the
product quality constraints, for example high quality of the copper cathodes.
There are several challenges associated with achieving these control objectives. Bergh
and Yianatos (2001) state that, to date, the lack of adequate mathematical models
applicable for industrial plants has prevented the development of control systems for
copper solvent extraction and electrowinning processes. However, development of the
dynamical models and the control system should not be too complicated. The
responses in extraction, with mixer-settler type of equipment, are sluggish and rather
stable due to the relatively low throughputs and large hold-up volumes (Wilkinson
and Ingham, 1983). The steady state effect of mass flow rate changes can be easily
calculated from the equilibrium scheme, for example a change in the aqueous to
organic flow rate changes the slope of the operating line in the McCabe Thiele
diagram, and subsequently also the output of the corresponding unit process.
Aqueous and organic entrainments, crud formation and phase separation times are
problems that could be solved with appropriate operating practices, separation
equipment, chemicals or with control strategies, as proposed by Bergh et al. (2006).
However, entrainments could pose a possible control problem if measurements of the
emulsion band and dispersion characteristics exist, and the phenomenon could be
modeled. In eletrowinning, problems such as operation of the baths are more of
maintenance/equipment development issues. (Bergh and Yianatos, 2001).
34/188
c(PLS) c(LE)
F(PLS) F(LE)
F(LO)
c(LO)
EXTRACTION STRIPPING
c(BO)
c(Raff) c(RE)
Figure 4-1: The solvent extraction process. The inputs are the copper concentration of the PLS,
c(PLS), and lean electrolyte, c(LE), and the flow rates of PLS, F(PLS), organic, F(LO), and lean
electrolyte, F(LE). The controlled variables are the copper concentration of the raffinate, c(Raff),
rich electrolyte, c(RE); and the copper concentrations of the recycled organic stream after
stripping, barren organic, c(BO), and after extraction, loaded organic, c(LO).
Table 4-1: Controlled, manipulated and disturbance variables of the copper solvent extraction
process.
corg
Extraction
equilibrium
isotherm
. .
c(LO)
Stripping Stripping
Extraction operating equilibrium
operating line line
c(BO) isotherm
. .
caq
Figure 4-2: The McCabe-Thiele diagram for the modeled process with one unit process for both
extraction and stripping, marked with dashed triangles. The input and output variables are
presented in Table 4-1. Extraction and stripping equilibrium isotherms and operating lines are
marked with solid lines. The horizontal axis is the copper concentration in the aqueous phase and
the vertical axis the copper concentration in the organic phase.
The equilibrium isotherm for extraction is nonlinear and for stripping it is linear, as
presented in Equations (3-1) and (3-2). In order to calculate the equilibrium output
copper concentrations from each unit process, the point where the equilibrium
isotherm and the inverse operating line overlap has to be determined. The calculation
is illustrated in Figure 4-3. The inverse operating line is first determined, and the
equation for the theoretical equilibrium point x* in the extraction is then derived. The
equilibrium points caq* and corg* for extraction are calculated on the basis of this
theoretical variable. The theoretical equilibrium point and the equilibrium points are
then derived for the stripping process and, finally, the steady state equations are
simplified and presented for the solvent extraction plant with one extraction unit and
one stripping unit.
37/188
corg
Extraction equilibrium isotherm
(x*,y*)
y* (caqout ,corgout)
corgout
Inverse extraction
operating line
caq
x* caq out α caq in
Figure 4-3: Calculation of the output concentrations (cout) for extraction with the nonlinear
extraction equilibrium isotherm and the efficiency coefficient α. The input copper concentrations
are marked with cin, and the equilibrium copper concentrations with ceq.
The slope of the inverse operating line is the negative ratio between the aqueous and
the organic flow rates to the mixer, -Faq / Forg. The inverse operating line is drawn
through the input copper concentrations (caqin, corgin). The inverse operating line is
determined as follows:
In the extraction, the theoretical equilibrium value for the aqueous copper
concentration, x*, is calculated by setting the equilibrium isotherm and the inverse
operating line (Equation 4-1) equal, as follows:
Ac aq
c org = = − F aq F org ⋅ c aq + ( cinorg + F aq F org ⋅ cinaq ) (4-2)
c +B
aq
For clarity, the coefficients of the inverse operating line are replaced with a and b, and
the second order equation is solved in relation to the aqueous copper concentration
caq, as follows:
Ac aq
c org = = ac aq + b
c +B
aq
⇒ c aq = 1
2a ( − ( Ba − A + b ) ± ( Ba − A + b )
2
− 4aBb )
The negative square root is chosen because the positive square root produces
unfeasible solutions (due to multiplication with 1/2a term, where a has negative value.
This follows from the assumption that flow rates are positive in Equation 4-2). The
coefficients of the inverse operating line, a and b, are replaced back to the original
variables. The theoretical equilibrium aqueous copper concentration is now:
38/188
x* =
2 F
−1
aq
F org
((
⋅ − − B ⋅ F F − A + ( cin + cin ⋅ F F )
aq org org aq aq org
)
(4-4)
( −B ⋅ F )
F org − A + ( cinorg + cinaq ⋅ F aq F org ) + 4 B ⋅ F aq F org ( cinorg + cinaq ⋅ F aq F org )
2
− aq
The output copper concentrations for the aqueous and organic phases are the
theoretical values weighted with the efficiency α:
org
cout = α i x * +(1 − α i ) ⋅ cinaq (4-5)
org
cout = −cout
aq
⋅ F aq F org + ( cinorg + cinaq ⋅ F aq F org ) (4-6)
For clarity, the extraction equilibrium isotherm is linearized around the operating
aq org
point (op1), represented by the aqueous copper concentration ( cop 1 , cop1 ). The slope of
the linearized curve is the derivate of the equilibrium curve evaluated at the
linearization point (Taylor series expansion). The linearized extraction equilibrium
isotherm is defined as follows:
A ( cop1)
aq 2
AB ) aq )
c org
= c aq
+ = Aic + Bi (4-7)
( B + cop
aq 2
1) ( B + copaq 2
1)
The theoretical equilibrium point is calculated by setting the inverse operating line
(Equation 4-1) equal to the linearized extraction equilibrium isotherm, and solving
this in relation to aqueous copper concentration caq. The theoretical equilibrium point
x* is now defined as follows:
) )
c org = Ai c aq + Bi = −cinaq ⋅ F aq F org + ( cinorg + cinaq ⋅ F aq F org )
)
⇒ x* =
( cinorg + cinaq ⋅ F aq F org ) − Bi
)
(4-8)
Ai + F aq F org
The output copper concentrations in the extraction, with the linearized equilibrium
isotherm curve, are formulated explicitly as a combination of the input concentrations.
The aqueous output copper concentration can be expressed by combining Equations
(4-4) and (4-7) as follows:
)
cinorg + cinaq ⋅F aq F org − Bi
c = αi
aq
) + (1 − α i ) ⋅ cinaq
Ai + F F
out aq org
) org ) (4-9)
aq (1 − α i ) Ai F + F aq org α i F org
α i Bi F org
= cin ) + cin ) org aq
− ) org aq
Ai F org + F aq Ai F + F Ai F + F
) aq ) org ) (4-10)
aq α i Ai F org Ai F + (1 − α i ) F aq α i Bi F aq
= cin ) org aq
+ cin ) + ) org aq
Ai F + F Ai F org + F aq Ai F + F
In stripping, the parameters of the linear equilibrium isotherm are C and D. The
theoretical equilibrium point x* is solved as in extraction. The inverse operating line
is set equal to the equilibrium isotherm, and the aqueous copper concentration x* is
solved as follows:
⇒ x* =
(c org
in + cinaq ⋅ F aq F org ) − D (4-11)
C + F aq F org
(4-12)
(1 − α i )CF org + F aq org α i F org α i DF org
= cinaq + cin org aq
− org aq
CF + F CF + F CF + F
org aq
(4-13)
The equilibrium state model for the plant is derived using these equations. The steady
state concentrations from each mixer-settler can be expressed as a function of the
incoming concentrations (cin), flow rates (F), stage efficiencies (α), and equilibrium
isotherm parameters (Ai,Bi,C,D). For the extraction step E, the aqueous and organic
output copper concentrations are derived from Equations (4-9) and (4-10) with the
variables listed in Table 4-1:
40/188
) )
(1 − α E ) AE F ( LO) + F ( PLS ) α E F ( LO) α E BE F ( LO)
c( Raff ) = c( PLS ) ) + c ( BO ) ) −
)
AE F ( LO) + F ( PLS ) AE F ( LO) + F ( PLS ) AE F ( LO) + F ( PLS )
(4-14)
) ) )
α A F ( PLS ) AE F ( LO ) + (1 − α E ) F ( PLS ) α E BE F ( PLS )
c ( LO ) = c ( PLS ) ) E E + c ( BO ) ) +
)
AE F ( LO ) + F ( PLS ) AE F ( LO) + F ( PLS ) AE F ( LO ) + F ( PLS )
(4-15)
For the stripping step S, the aqueous and organic output copper concentrations are
derived from Equations (4-12) and (4-13) using the variables listed in Table 4-1:
α S CF ( LE ) CF ( LO ) + (1 − α S ) F ( LE ) α S DF ( LE )
c( BO) = c( LE ) + c( LO) + CF ( LO ) + F ( LE )
CF ( LO) + F ( LE ) CF ( LO) + F ( LE )
(4-17)
The steady state of the system can be solved from these equations by substitution.
Estimation of the equilibrium isotherm parameters, the efficiencies and recycle
corrections are explained in more detail in the following sections.
[CuR2 ] + [ RH ] = [ R ] = constant
2 2
(4-18)
[ RH ] = [ R ] − [CuR2 ]
2 2
(4-19)
Assume that the concentration of [CuR2] can be approximated by the organic copper
concentration corg and the copper ion concentration by the aqueous copper
41/188
concentration caq. Now by replacing Equation (4-19) to the equilibrium Equation (2-
3), the following equation for the equilibrium constant is formulated:
[CuR2 ] ⋅ H + c org ⋅ c ( H + )
2 2
κ= ≈ aq
Cu +2 ⋅ ([ R ] − [CuR2 ]) c ⋅ ( c( R) − c )
2 2 org
(4-20)
and by rearranging this equation we get the form of the extraction equilibrium
isotherm with the extraction equilibrium constant κE :
κ E ⋅ c( R) 2 ⋅ c aq c( R) 2 ⋅ c aq
c org = = (4-21)
c( H + ) 2 + κ E ⋅ c aq c( H + ) 2 / κ E + c aq
For the stripping equilibrium isotherm the equation is the same with the stripping
equilibrium constant κS but, since the acidity is much higher, the hydrogen ion
concentration becomes the determining term, and the equation simplifies into a linear
form with the estimation error term ε:
κ S ⋅ c( R )2 ⋅ c aq κ S ⋅ c( R) 2 ⋅ c aq
c org = ≈ +ε (4-22)
c( H + )2 + κ S ⋅ c aq c( H + ) 2
cinaq − cout
aq
cinorg − cout
org
αE = = (4-23)
cinaq − x cinorg − y
Calculation of the efficiency for the extraction unit is presented in Figure 4-3. The
output concentrations cout are determined by drawing an inverse operating line from
the input concentrations cin towards the extraction equilibrium isotherm. The point
where this line and the equilibrium isotherm overlap is the theoretical equilibrium
value for the aqueous X and organic Y concentrations. These values are then weighed
with the efficiency parameter α to obtain the output values cout. The efficiency
parameter typically has values close to 1; values above 1 are possible for real plants
because the efficiency is a theoretical measure. (Ingham et al., 1994)
4.2.3 Estimation of the recycle corrections
It is often possible within the mixer-settler units to recycle one of the phases in order
to change the aqueous to organic phase ratio without changing the flow rates. If the
recycle flow rates are not measured, then the aqueous to organic flow ratios in the
mixer have to be estimated from the steady state data of the plant. The recycle
correction is based on the mass balance of copper in the mixer. The aqueous copper
42/188
mass is multiplied by the recycle correction coefficient, cf, to equal with the organic
copper mass as follows:
cf ⋅ F aq ( cinaq − cout
aq
) = − F org ( cinorg − coutorg ) (4-24)
The correction coefficient can now be estimated from the steady state data of the plant
as follows:
− F org ( cinorg − cout
org
)
cf = (4-25)
F aq ( cinaq − cout
aq
)
4.3 Dynamic models of the copper solvent extraction process
A mechanistic model approach was preferred for the dynamic model in order to be
able to describe the detailed process phenomena. The dynamic model development
was started from the basic formulation for mixer-settlers presented in Wilkinson and
Ingham (1983). This approach was chosen due to the successful earlier studies
described in Chapter 3. The dynamic models are based on ideal mixing and plug flow
models. In this work, the ideal mixing model is modified to utilize the equilibrium
values calculated with the equilibrium state model. The assumptions used by
Wilkinson and Ingham (1983) that are applicable to the developed dynamic model
are: (1) perfect mixing in the mixer, (2) immiscibility of the two phases, and (3) no
mass transfer or back-mixing in the settler (plug flow). In this work, the following
assumptions are added: (4) the equilibrium curve in the mixer is a plant specific non-
linear equilibrium isotherm, (5) the mass transfer coefficients Ki, the equilibrium
isotherm parameters A and B for extraction, and C and D for stripping, and the
efficiency coefficients αi, are not constant, but estimated from the offline plant
measurements. (6) The phase volumes are calculated on the basis of the phase flow
rates and total equipment volumes, and (7) hydrodynamic effects are neglected. (8)
The equilibrium and the phase ratio in the mixer are reached gradually, not
immediately as in Wilkinson and Ingham (1983).
The copper mass transfer is calculated from the ideal mixing Equation (3-6), where
the equilibrium value c* is determined on the basis of the equilibrium state model
described in the previous section. The variables are marked as follows: flow rates F,
the mixing volumes Vm, organic concentrations corg, aqueous concentrations caq, mass
transfer coefficients Ki, efficiency parameters αi. The settler, which always follows the
mixer, is described by a pure time delay ti .
In extraction, copper is transferred from the aqueous to the organic phase. The
extraction unit operation is modelled by differential equations of the concentrations
for both the organic ( dc1org (t ) dt ) and aqueous phases ( dc1aq (t ) dt ), where the
equilibrium output value c* is calculated from the equilibrium state model:
43/188
dc1org (t ) F1org (t )
= ⋅ c0org (t − t0 ) − c1org (t ) + K1 c1org (t ) − c1org * (t ) (4-26)
dt Vm
dc1aq (t ) F1aq (t )
= ⋅ c0aq (t ) − c1aq (t ) − K1 c1org (t ) − c1org * (t ) (4-27)
dt Vm
where the incoming copper concentrations are marked with subscript 0 and the copper
concentrations in the mixer with subscript 1. The settler is taken into account by
delaying the mixer output concentrations for the aqueous phasewith a time delay t1
and for the organic phase with a time delay t2:
In stripping, copper is transferred from the organic to the electrolyte solution. The
stripping unit operation is modeled by differential equations of the concentrations for
both electrolyte ( dc1el (t ) dt ) and organic ( dc2org (t ) dt ) phases, taking into account the
equilibrium concentration c* calculated from the equilibrium state model:
dc1el (t ) F1el (t )
= ⋅ c0el (t ) − c1el (t ) − K 2 c1el (t ) − c1el * (t ) (4-31)
dt Vm
dc2org (t ) F2org (t )
= ⋅ c1org (t − t2 ) − c2org (t ) − K 2 c1el (t ) − c1el * (t ) (4-32)
dt Vm
where the incoming electrolyte copper concentrations are marked with subscript 0,
and the electrolyte copper concentrations in the mixer with subscript 1. Since the
organic solution comes from the extraction unit, the incoming organic copper
concentrations are marked with subscript 1 and the organic copper concentrations in
the mixer with subscript 2. Rich electrolyte concentration (c(RE)) is the time delayed
(t3) value of the electrolyte concentration, and barren organic (c(BO)) is the time
delayed (t4) value of the organic concentration in the stripping unit.
Since the organic solution is recycled from the stripping unit back to the extraction
unit, c(BO) is c0org and t4 is same as t0 in Equation (4-26) and Equation (4-28).
44/188
These four equations (4-26, 4-27, 4-31, 4-32) are ordinary differential equations that
can be solved with standard techniques. These time variant models have four states,
copper concentrations of aqueous and organic phases in both extraction and stripping
units.
4.3.1 Calculation of the mass transfer coefficients
The mass transfer coefficient K describes the speed of the concentration change in the
mixers. It is inversely related to the process time constant: the higher the K value, the
shorter the time constant. The mass transfer coefficient K is approximated from the
process model by setting it to a steady state and giving a small value (a tuning
coefficient 0.01 was found that gave appropriate time dynamics) to the difference
term (cout − cout *) .
dcout
V = F (cin − cout ) − K (cout − cout *)V = 0
dt
(4-36)
F (cin − cout ) F (cin − cout )
⇔K= =
V (cout − cout *) V ⋅ε
Using this procedure the K values are between 50 – 400 1/s for the studied process.
Wilkinson and Ingham (1983) suggested giving an arbitrary high value for the K·V
term. The modelling approach in Ingham et. al. (1994) is slightly different; they use a
constant mass transfer coefficient K = 25 1/s.
4.3.2 Estimation of the mixer and settler phase volumes
The mixing model includes the mixer volume, Vm, and the settling model settler
volume, Vs, both for the aqueous and organic phases. Since there are no measurements
of the volumetric ratio between the organic and the aqueous phases in the mixer and
the settler, the ratio is assumed to follow the incoming flow ratio to the mixer, as
suggested by Ingham et al. (1994, p 186). Considering that V represents the total
volume of the equipment, the volume of the organic phase can be represented as:
F org
V org = V (4-37)
F org + F aq
and the volume of the aqueous phase as:
F aq
V aq = V (4-38)
F org + F aq
4.3.3 Estimation of the time delay for the plug flow model
The plug flow model describing the settler has one parameter, a time delay t. The
organic mean residence time (delay) is the total organic volume divided by the
organic flow rate, which simplifies into the total volume divided by the total flow
rate:
45/188
F org
org Vs ⋅
t org =
V
= F org + F aq = Vs
(4-39)
F org F org
F + F aq
org
and the aqueous mean residence time (delay) is the total aqueous volume divided by
the aqueous flow rate, simplifying into the same form as the organic time delay:
F aq
Vs ⋅
V aq F org + F aq = Vs
t aq = = (4-40)
F aq F aq
F + F aq
org
Thus, the time delay for the aqueous phase and organic phase in the settler are equal.
The result is justified by the operating practice of the case industrial copper solvent
plant: The flow rates organic and aqueous phases in the settler are maintained on
equally levels in order to minimize problems on the organic/aqueous interphase layer
such as entrainment and thickening of the emulsion layer.
46/188
Organic solution
c(RaffP) c(RaffS) c(RE)
Figure 5-1: The copper solvent extraction process. The output variables are the copper
concentrations of the raffinates c(RaffP) and c(RaffS), rich electrolyte, c(RE), and the copper
concentrations of the recycled organic stream, barren organic, c(BO) and loaded organic c(LO).
The input variables are the flow rates of PLS, F(PLSS) and F(PLSP), organic, F(LO), and lean
electrolyte, F(LE), and the copper concentration of PLS, c(PLS) and lean electrolyte, c(LE).
The online measurements consist of two organic and five aqueous copper
concentrations and four flow rates. The offline measurements include pH of the PLS
solution (pH), acidity of the electrolyte solution (acid), the reagent volume per cent in
the organic solvent (vol) (for description of the reagent, see Section 2.3.3). The offline
measurements also include four organic and five aqueous copper concentrations. The
measurements with information about the measurement type (online/offline) are listed
in Table 5-1.
In the copper solvent extraction and electrowinning plant there are no advanced
control systems. The process operators change the flow rates in the solvent extraction
process and current amperages in the electrowinning process in order to keep the lean
electrolyte copper concentration within the target value. The lean electrolyte copper
concentration target value is set in order to maximize the production of copper
cathodes by the plant management. The control structure and strategy are further
discussed in Chapter 9.
A preliminary classification of the controlled, manipulated and disturbance variables
of the copper solvent extraction plant was performed on the basis of the operational
and process knowledge. The possible controlled variables are the outputs of the
extraction and stripping processes; rich electrolyte concentration, c(RE), raffinate
concentrations c(RaffS) and c(RaffP), and organic concentrations c(LO) and c(BO).
The manipulated variables are the four flow rates of the leach, organic and electrolyte
solutions. The measured disturbance variables are the leach and electrolyte solution
concentrations. This classification is presented in Table 5-1.
48/188
Table 5-1: Controlled, manipulated, disturbance and state variables of the industrial copper
solvent extraction process. The measurement type, online/offline, is indicated on fourth and fifth
column, with an indication of the online and offline measurements.
∑( y − yi )
N
1 2
aSSE = m ,i (5-1)
N i =1
The average sum of squared errors for the online measurements are presented in Table
5-2. The averages of the squared error sums are higher for the variables with higher
values. The squared errors between the measurements and filtered signals are
considerable for all the variables with all the filtering periods. This suggests that there
is significant measurement noise, and filtering is therefore necessary.
Table 5-2: Average sum of squared errors for the online measurements with different averaging
periods (3, 6, 12, 18, 24 and 30 sampling times).
period 3 6 12 18 24 30
c(PLS) 0.1606 0.1609 0.1612 0.1614 0.1615 0.1617
c(RaffS) 0.0065 0.0065 0.0066 0.0066 0.0067 0.0067
c(RaffP) 0.0010 0.0011 0.0011 0.0011 0.0011 0.0011
c(LE) 20.2950 20.2985 20.3151 20.3113 20.2982 20.2882
c(RE) 35.8249 35.8738 35.9320 35.9615 35.9863 36.0086
c(LO) 2.6428 2.6464 2.6477 2.6505 2.6553 2.6608
c(BO) 0.2534 0.2535 0.2536 0.2538 0.2539 0.2540
F(PLSS) 0.5618 0.5665 0.5716 0.5744 0.5774 0.5805
F(PLSP) 0.5110 0.5225 0.5304 0.5347 0.5381 0.5412
F(LE) 0.1329 0.1396 0.1537 0.1648 0.1739 0.1823
F(LO) 0.3092 0.3117 0.3160 0.3198 0.3237 0.3276
For the PLS copper concentration an averaging period of 6 sample times or less gave
relatively noisy signals, whereas the period of 30 sample times removed too much
variation from the data for modeling purposes. An averaging period of 12 sample
50/188
times produced low residuals (aSSE 0,16), and visual examination of the data
confirmed that the filtered signal followed smoothly the raw data. As can be seen
from Figure 5-2 for the PLS copper concentration, the dynamic changes in the data
are still present and the noise has been removed. The PLS copper concentration
residuals are approximately white noise, as can be seen from Figure 5-3.
Figure 5-2: The original data with the outliers (dotted, red) and filtered data (solid, black)
measurements of the PLS copper concentration.
Figure 5-3: Residuals between the original measurement data with the outliers and filtered
measurements of the PLS copper concentration.
algorithm was chosen for the bias correction. The algorithm with a forgetting factor of
0.95 was used to estimate the bias coefficient b between the online y(t) and offline u(t)
measurements with the recursive least squares method, as follows:
y (t ) = b ⋅ u (t ) + e(t ) (5-2)
Figure 5-4: The original PLS online concentration measurement (dotted, red), bias corrected
concentration measurements (solid, black) and the laboratory measurements (dashed, blue).
points in the neighbouring clusters. Points that are very distant from the neighbouring
clusters have the value 1. If a point is not especially close to any of the clusters, the
silhouette value is –1.
On the basis of the distance sums and cluster means plotted into the process data, the
cases with 3,6 and 9 were examined in more detail. The corresponding silhouettes are
illustrated in Figure 5-5, Figure 5-6 and Figure 5-7. The case with three clusters
separates one extreme point and clusters the rest of the data points into two well
separated large clusters. The case with six clusters separates two extreme points,
while clustering the rest of the points into three smaller groups and one well separated
large group. The case with nine clusters separates three extreme points, one large
group and five smaller groups, two of which are not well separated.
The cluster centers were illustrated by plotting them into the process data, as shown in
Figure 5-8 for the six-cluster case. Three clusters were not enough to cover the
different operating points, whereas with 9 clusters some of the cluster centres were
almost the same. With six clusters the four normal and two of the extreme operating
points were covered well, as illustrated in Figure 5-8 and Figure 5-9. The distances
between the closest data points and the mathematical cluster centres were small. Only
the two normal operating points from the start of each data set (first and fourth circle
in both Figure 5-8 and Figure 5-9 were considered for further study. In the following,
these two operating points are called the first operating point (DP1) and the second
operating point (DP4).
54/188
Figure 5-8: The flow rates of the PLS parallel, PLS series, lean electrolyte and loaded organic.
The data points closest to the cluster centres are circled. The first and fourth circles (around
samples 2 and 46) are operating points DP1 and DP4 chosen for further study.
Figure 5-9: Copper concentrations of the PLS, barren organic, loaded organic, lean electrolyte
and rich electrolyte. The data points closest to the cluster centres are circled. The first and fourth
circles (around samples 2 and 46) are operating points DP1 and DP4 chosen for further study.
55/188
i+ p
1 n 1 n 1
Variation = ⋅ ∑ yi − yi = ⋅ ∑ yi − ∑ yj (5-3)
n i =1 n i =1 2 p + 1 j =i − p
where y is the filtered online measurement at time (i), n is the number of data points in
one week, and p is the number of data points in one day. The variation index is scaled
by dividing it with the nominal value of the variable.
The data set included 18 weeks of operational online measurements of the controlled,
manipulated and disturbance variables. The variations in the possible controlled
variables (CV), manipulated variables (MV) and disturbance variables (DV) are
presented in Table 5-3.
Table 5-3: Variations of the online measured controlled (CV), manipulated (MV) and
disturbance (DV) variables.
CV CV CV CV CV MV MV MV MV DV DV
c(RE) c(LO) c(RaffS) c(RaffP) c(BO) F(PLSS) F(PLSP) F(LE) F(LO) c(PLS) c(LE)
% % % % % % % % % % %
W1 1.30 3.20 15.30 10.31 2.92 1.99 3.04 1.25 1.77 2.30 1.58
W2 0.79 0.56 1.87 3.54 0.75 1.35 1.14 1.87 1.10 0.92 1.00
W3 0.84 1.53 7.64 3.30 0.69 1.06 0.95 1.12 0.65 3.57 1.16
W4 1.29 1.80 9.33 6.18 2.41 2.67 2.68 2.24 1.31 2.61 1.52
W5 0.97 2.45 9.03 6.54 3.03 0.90 1.22 2.47 0.95 2.37 1.18
W6 1.10 2.84 8.99 9.25 2.72 0.50 0.49 0.61 1.00 0.78 1.47
W7 0.88 1.98 6.36 5.15 1.96 0.83 0.43 0.38 0.84 0.87 0.95
W8 0.82 1.45 11.47 9.23 2.62 0.31 0.20 1.81 0.85 0.89 0.96
W9 0.93 1.44 8.21 9.62 1.78 0.11 0.17 2.72 0.75 0.87 1.47
W10 0.86 1.18 6.54 7.55 2.06 1.06 0.36 2.28 0.96 0.95 1.13
W11 1.35 3.00 8.15 7.33 2.01 0.59 0.34 2.80 0.75 0.84 1.71
W12 1.19 2.35 10.12 12.62 2.22 2.69 0.65 3.11 1.52 1.37 1.71
W13 1.18 0.70 6.62 7.47 1.68 1.50 0.25 2.62 1.05 1.39 1.81
W14 0.84 1.62 5.91 5.47 2.49 0.54 0.42 2.11 0.93 1.04 1.01
W15 2.44 1.30 13.02 9.48 1.93 1.64 1.54 2.73 1.10 1.13 3.09
W16 1.51 0.96 9.32 7.82 2.62 0.67 1.71 1.38 1.10 0.93 2.29
W17 1.59 1.22 17.06 10.47 2.58 1.55 2.38 2.63 1.05 1.78 2.64
W18 1.42 0.93 7.69 6.75 2.48 0.50 0.41 2.94 0.45 0.67 2.27
Min 0,79 0,56 1,87 3,30 0,69 0,11 0,17 0,38 0,45 0,67 0,95
Max 2,44 3,20 17,06 12,62 3,03 2,69 3,04 3,11 1,77 3,57 3,09
Mean 1,18 1,70 9,04 7,67 2,16 1,14 1,02 2,06 1,01 1,40 1,61
The maximum variation in the rich electrolyte, c(RE), loaded organic, c(LO) and
barren organic c(BO) copper concentrations was between 2,5 – 3,2 %. The variations
56/188
in the raffinate copper concentrations, c(RaffS) and c(RaffP), are significantly higher
(between 12 – 17%) than for the other controlled variables. The average variation in
the raffinate copper concentrations is 7,7 – 9,0%, which is very large compared to that
for the other variables. This might imply that the raffinate copper concentration
measurements are less accurate than the other copper concentration measurements,
and thus should not be used as controlled variables in the further studies.
The maximum variations for the manipulated variables (MV), the PLS series and
parallel, organic and electrolyte flow rates, were around 2,8 – 3,1%. The largest
average variation was in the electrolyte flow rate F(LE) and the smallest average
variations in the organic flow rate F(LO) and PLS parallel flow rate F(PLSP). The
disturbance variables (DV), PLS and lean electrolyte copper concentrations, had
maximum variations of between 3,1 – 3,6%. On the average the variations were
between 1,4 – 1,6%.
The following suggestions for the modeling and control studies are formulated on the
basis of this analysis; the maximum realistic changes for the controlled and
disturbance variables are less than ±5% around the chosen operating point. The
maximum changes in the manipulated variables are less than ±3% around the chosen
operating point. This study also suggests that the raffinate copper concentrations
should not be used as controlled variables, since the large variations in these variables
might indicate serious measurement inaccuracy. Variation starting from 0,5 % in the
rich electrolyte copper concentration is significant for the production of the copper
cathodes.
5.2.4 Remarks on the quality of the process data
Based on the preliminary analysis of the available online and offline data from the
case industrial copper solvent extraction plant, the remarks are given on the process
data quality.
Most of the online copper concentration measurements and the flow rate
measurements have some noise, but after the data filtering, these measurements can
be considered to be reliable. The online raffinate and loaded organic copper
concentration measurements have high noise level and at some periods possible
measurement inaccuracies. Therefore, the interpretation of these three variables
should be done carefully.
The offline measurements of the intermediate copper concentrations, pH, acidity and
reagent volume percent are considered to be reliable. The problem with pH, acidity
and reagent volume percent measurements is the long sampling interval.
The internal recycle flow rates in the mixer-settler units are adjusted with manual
valves, and thus these are not measured on the plant. Therefore in this study, these
recycle flow rates have to be approximated.
57/188
) )
(1 − α EP ) AEP F ( LO) + F ( PLSP) α EP F ( LO) α EP BEP F ( LO)
c( RaffP) = c( PLS ) ) + c( BO) ) − )
AEP F ( LO) + F ( PLSP) AEP F ( LO) + F ( PLSP) AEP F ( LO) + F ( PLSP)
(6-1)
) ) )
α EP AEP F ( PLSP ) AEP F ( LO ) + (1 − α EP ) F ( PLSP) α EP BEP F ( PLSP)
c ( BO1) = c( PLS ) ) + c ( BO ) ) + )
AEP F ( LO) + F ( PLSP) AEP F ( LO ) + F ( PLSP) AEP F ( LO ) + F ( PLSP)
(6-2)
For the first series extraction unit, E1, the aqueous and organic equilibrium copper
concentrations are:
) )
(1 − α E 2 ) AE 2 F ( LO) + F ( PLSS ) α E 2 F ( LO) α E 2 BE 2 F ( LO)
c( RaffS ) = c( PLS1) ) + c( BO1) ) − )
AE 2 F ( LO) + F ( PLSS ) AE 2 F ( LO) + F ( PLSS ) AE 2 F ( LO) + F ( PLSS )
(6-3)
58/188
) ) )
α A F ( PLSS ) AE 2 F ( LO) + (1 − α E 2 ) F ( PLSS ) α E 2 BE 2 F ( PLSS )
c( BO 2) = c( PLS1) ) E 2 E 2 + c ( BO1) ) +
)
AE 2 F ( LO) + F ( PLSS ) AE 2 F ( LO) + F ( PLSS ) AE 2 F ( LO) + F ( PLSS )
(6-4)
For the second series extraction unit, E2, the aqueous and organic equilibrium copper
concentrations are:
(1 − α ) A) F ( LO) + F ( PLSS ) α E1 F ( LO ) )
α E1 Bi F ( LO)
c( PLS1) = c( PLS ) )
E1 E1
+ c ( BO 2) ) − )
AE1 F ( LO ) + F ( PLSS ) AE1 F ( LO ) + F ( PLSS ) AE1 F ( LO) + F ( PLSS )
(6-5)
) ) )
α E1 AE1F ( PLSS ) AE1F ( LO) + (1 − α E1 ) F ( PLSS ) α E1BE1F ( PLSS )
c( LO) = c( PLS ) ) + c( BO 2) ) + )
AE1F ( LO) + F ( PLSS ) AE1F ( LO) + F ( PLSS ) AE1F ( LO) + F ( PLSS )
(6-6)
For the stripping unit, S, the aqueous and organic equilibrium copper concentrations
are:
α S CF ( LE ) CF ( LO ) + (1 − α S ) F ( LE ) α S DF ( LE )
c( BO) = c( LE ) + c( LO) + CF ( LO ) + F ( LE )
CF ( LO) + F ( LE ) CF ( LO) + F ( LE )
(6-8)
Calculation of all the copper concentrations in the steady state requires iteration due
to organic solvent recycle between the extraction and stripping units [c(BO) - c(BO1)
- c(BO2) - c(LO) - c(BO)], and the aqueous recycle between series extraction units E1
and E2 [c(PLS) – c(PLS1) – c(RaffS)]. The steady state equilibrium copper
concentrations of the solvent extraction plant can be evaluated using this set of eight
equations.
For the parallel extraction unit, EP, the aqueous and organic copper concentrations
after the mixer are determined as follows:
dc1org (t ) F1org (t )
= ⋅ c0org (t − t0 ) − c1org (t ) + K1 c1org (t ) − c1org * (t ) (6-9)
dt Vmix ,1 (t )
dc1aq (t ) F1aq (t )
= ⋅ c0aq (t ) − c1aq (t ) − K1 c1org (t ) − c1org * (t ) (6-10)
dt Vmix ,1 (t )
where the equilibrium value is calculated from the equilibrium state model with the
following variables:
For the first series extraction unit, E1, the aqueous and organic copper concentration
after the mixer are determined as follows:
dc2org (t ) F2org (t )
= ⋅ c1org (t − t1 ) − c2org (t ) + K 2 c2org (t ) − c2org * (t ) (6-12)
dt Vmix ,2 (t )
dc2aq (t ) F2aq (t )
= ⋅ c1aq (t ) − c2aq (t ) − K 2 c2org (t ) − c2org * (t ) (6-13)
dt Vmix ,2 (t )
where the equilibrium value is calculated from the equilibrium state model with the
following variables:
For the second series extraction unit, E2, the aqueous and organic copper
concentration after the mixer are determined as follows:
dc3org (t ) F3org (t )
= ⋅ c2org (t − t2 ) − c3org (t ) + K 3 c3org (t ) − c3org * (t ) (6-15)
dt Vmix ,3 (t )
dc3aq (t ) F3aq (t )
= ⋅ c2aq (t ) − c3aq (t ) − K 3 c3org (t ) − c3org * (t ) (6-16)
dt Vmix ,3 (t )
where the equilibrium value is calculated from the equilibrium state model with the
following variables:
The parallel raffinate copper concentration, c(RaffP), after the EP settler is the time
delayed value of the aqueous copper concentration from the EP mixer.
60/188
The series raffinate copper concentration, c(RaffS), after the E1 settler is the time
delayed value of the aqueous copper concentration from the E1 mixer.
The loaded organic copper concentration, c(LO), after the E2 settler is the time
delayed value of the organic concentration from the E2 mixer.
In the stripping unit, copper is transferred from the organic to the electrolyte solution.
The stripping unit operation is modelled using differential equations of the
concentrations for both the electrolyte ( dciaq (t ) dt ) and organic ( dciorg (t ) dt ) phases
as follows:
dc1el (t ) F1el (t )
= ⋅ c0el (t ) − c1el (t ) + K 4 c1el (t ) − c1el* (t ) (6-21)
dt Vmix ,4 (t )
dc4org (t ) F4org (t )
= ⋅ c3org (t − t3 ) − c4org (t ) − K 4 c1el (t ) − c1el* (t ) (6-22)
dt Vmix ,4 (t )
where the equilibrium copper concentration is calculated from the equilibrium state
model with the following variables:
The rich electrolyte concentration (c(RE)) is the time delayed value of the electrolyte
concentration, and the barren organic concentration (c(BO)) is the time delayed value
of the organic concentration in the stripping unit.
The organic tank and wash stage between the extraction and stripping stages are
modeled as a lumped time delay.
Extraction
c(PLS), F(PLSP)
E1P
c(RaffP), F(RaffP)
c(BO1)
F(BO1)
Extraction
E1S
c(PLS1)
F(PLS1) c(RaffS), F(RaffS)
c(BO2)
F(BO2)
Extraction
c(PLS), F(PLSS)
E2S
c(LO)
F(BO3)
Wash&Tank
F(LO) c(LO)
Stripping
c(LE), F(LE) c(BO)
S1H
F(BO)
c(RE), F(RE)
Figure 6-1: Simulation model of the copper solvent extraction process. The parallel extraction
unit is marked with E1P, the first series extraction unit with E1S, and the second extraction unit
with E2S. The stripping unit is marked with S1H and the organic storage tank with ‘Tank’. The
input variables (left) are the copper concentrations c(PLS) and c(LE) and the flow rates F(PLSP),
F(PLSS), F(LO), F(LE). The state variables are the intermediate copper concentrations c(BO),
c(BO1), c(BO2), c(LO), c(PLS1) marked with italics. The output variables (right) are the copper
concentrations c(RE), c(RaffS) and c(RaffP).
The inputs of the simulator are the copper concentrations of the leach and electrolyte
solutions, c(PLS) and c(LE), and the flow rates of the parallel and series leach
solutions, the organic solution and the electrolyte solution, F(PLSP), F(PLSS), F(LO),
F(LE). The intermediate copper concentrations are the partial leach solution c(PLS1)
from the second extraction step, the barren organic concentration c(BO) from
stripping, the first partial organic concentration c(BO1) from the parallel extraction
step, second partial c(BO2) organic concentration from first series extraction step, and
the loaded organic concentration c(LO) from the second series extraction step. The
62/188
loaded organic copper concentration is then led to the tank model, which is a pure
time delay, and then to the stripping stage. The output copper concentrations from the
model are the raffinate parallel and series concentrations, c(RaffP) and c(RaffS), and
the rich electrolyte concentration c(RE). The input variables, state variables, output
variables, parameters and constants of the simulation model are listed in Table 6-1.
Table 6-1: Input variables, state variables, output variables, parameters and coefficients of the
simulation model.
Parameters
Eq. isotherm A,B,C,D
MIXER
Efficiency α
Mass transfer of copper
Recycle corr. cf
Equilibrium copper
Mass transfer coef. K
concentration c*
Mixer Volume Vm
coutmorg coutmaq
Foutmorg Foutmaq
SETTLER SETTLER
Organic delay Aqueous delay
coutsorg coutsaq
Foutsorg Foutsaq
mixer model are delayed according to Equation 6-24, where the time delay is
calculated on the basis of Equation 4-38.
cinmaq
cinmorg
McCabe-Thiele parameters
Equilibrium calculation A/O ratio Recycle
Eq. isotherm A,B,C,D
caq* cf• Finmaq correction cf
Efficiency α
Eq.(4-10) Finmorg Eq. 4-25
Eq. 4-22, Eq. 4-23
Equilibrium dynamics
1
(bs+1)
Mass transfer
Mass transfer coef. K Vmixorg Mixer
dcorg/dt, dcaq/dt
Eq. 4-36 Eq. 4-37 Volume Vm
Eq.6-21 , Eq. 6-22
coutsorg coutsaq
Figure 6-3: Detailed structure of the mixer-settler model with equations for the stripping unit.
The input variables and parameters are marked with dashed rectangles. Calculations are
marked with solid rectangles.
The inputs and outputs of each mixer-settler unit model are listed in Table 6-2, and
the equations of the parameter estimation and the dynamic calculations in Table 6-3.
Table 6-2: The inputs and outputs of each mixer-settler unit model.
Table 6-3: The equations of the parameter estimation and dynamic calculation of each mixer-
settler unit model.
Figure 7-1: PLS and lean electrolyte copper concentrations, reagent volume per cent in the
organic solution, pH of the PLS solution and acidity of the electrolyte solution.
Figure 7-2: Flow rates of the PLS series, PLS parallel, organic solution and electrolyte solution.
The online copper concentrations, extrapolated offline measurements and flow rates
of the second data set are presented in Figure 7-3 and Figure 7-4. This data set is
characterized by a slow rise in both the PLS copper concentration and reagent volume
68/188
per cent in the organic solution after sampling time 1000. The lean electrolyte
concentration change considerably during the whole period, and there is a decrease in
the flow rates around sampling period [2200 – 2500].
Figure 7-3: PLS and lean electrolyte copper concentrations, reagent volume per cent in the
organic solution, pH of the PLS solution and acidity of the electrolyte solution.
Figure 7-4: Flow rates of the PLS series, PLS parallel, organic solution and electrolyte solution.
69/188
Ac aq c( R ) 2 ⋅ c aq φ (vol ) 2 ⋅ c aq
c org = = =
B + c aq c( H + ) 2 K e + c aq ϕ (10− pH ) 2 + c aq
(7-1)
For the modeling there are 144 data points from 18 simulated equilibrium isotherms,
and for validation 2 data sets with 16 data points. The variables are the initial aqueous
and organic concentrations (x0 and y0), the reagent volume percent in the organic
solution (vol), the pH of the solution, and the equilibrium values for the aqueous and
organic concentrations. In order to get a good fit for the equilibrium isotherm over the
whole operating range, the criterion applied in model fitting is minimization of the
absolute error sum between the measurements and model predictions.
The following model structure describes adequately well the extraction equilibrium
data:
φ (vol ) ⋅ c aq
c org = (7-2)
ϕ ( H + ) 2 (c org 0 / c aq 0 ) + c aq
The constant parameters are [φ=0.51 ϕ=253.80]. The model fit is good, the absolute
error sum is 2.77, and measurement and the predicted points overlap very well, as can
be seen from Figure 7-5.
70/188
Figure 7-5: Validation datapoints (o) and predicted data points (*) of the extraction equilibrium
isotherm.
The basic structure for the stripping equilibrium isotherm consists of squares of the
reagent volume per cent (vol) and acidity (acid) with the reagent specific constant φ
and the inverse reaction constant ϖ, and the estimation error term ε. The stripping
equilibrium isotherm is defined as follows:
c( R ) 2 ⋅ c aq φ (vol ) 2 ⋅ c aq
c org = C ⋅ c aq + D = + ε = +ε (7-3)
c( H + ) 2 K S ω (acid ) 2
The simulated modeling data for the stripping equilibrium isotherm include 165 data
points, with 24 data points close to the normal operating conditions. The model
validation data include two real isotherms measured in the plant, including a total of 6
data points. The variables are the initial concentration of the aqueous solution (x0), the
reagent volume per cent in the organic solution (vol), the acidity of the solution (acid),
and the equilibrium values for the aqueous and organic concentrations (x and y).
The following model structure with the parameter values [φ/ϖ=0.48, ψ=0. 75, ζ=-
0.04] describes adequately well the stripping equilibrium data:
ψ
φ (vol ) aq
c org
= ⋅ c + ζ x0
ω (acid )
(7-4)
The model has adequate fit to the measurements, as can be seen from Figure 7-6.
71/188
Figure 7-6: Validation data points (o) and predicted data points (*) of the stripping equilibrium
isotherm.
Figure 7-7: Scaled values of the varied equilibrium isotherm parameters: A and B for extraction,
and C and D for stripping for the first test period.
The changes in the equilibrium isotherm parameters in the second test period are
shown in Figure 7-8. In the figure the isotherm parameter A increases significantly
during the test period, leading to a higher extraction potential of copper from the PLS
solution to the organic solution.
Figure 7-8: Scaled values of the varied equilibrium isotherm parameters: A and B for extraction,
and C and D for stripping for the second test period.
Both of the two available offline measurement data sets included around 35 data
points. The plant operation was assumed to be stable, with small daily variation. As a
result it was therefore also assumed that the offline copper concentration
measurements, average flow rates and estimated equilibrium isotherms would
represent the steady state of the process.
In the varied parameter approach, the efficiency and recycle correction parameters are
calculated directly on the basis of the offline measurements, and no optimization is
necessary to fit the parameters.
In the constant parameter approach, optimization is necessary to fit the parameters to
the offline data sets. The optimization for the parameter estimation is explained in
detail in the following.
An optimization model is constructed on the basis of the equilibrium state models
presented in Section 6.1. The parameter estimation is based on minimizing the error
between the equilibrium state model outputs and the measurements, listed in Table
7-1. If the steady state assumption is valid, then with optimal parameters all the error
measures approach zero.
Table 7-1: Inputs, outputs, estimated parameters and error measures for all four unit processes.
It is assumed that the parameters are different for the two offline data sets. Therefore,
both are optimized separately with the Matlab lsqcurvefit algorithm for nonlinear
curve-fitting. The cost function aims to minimize the error measures presented in
Table 7-1. In order to minimize the recycle loop effects, the errors between the offline
measured and estimated copper concentrations of the barren organic (c(BO)) and
partial PLS (c(PLS1)) concentrations are weighted more in the cost function than the
errors between the offline measured and estimated copper concentrations of the rich
electrolyte (c(RE)) and loaded organic (c(LO)). The cost function is defined as
follows:
N N
Cost = 10∑ c( BO)meas (i ) − c( BO)est (i ) + 10∑ c( PLS1) meas (i ) − c( PLS1)est (i )
i =1 i =1
(7-5)
N N
+ ∑ c( RE )meas (i ) − c( RE )est (i ) + ∑ c( LO)meas (i ) − c( LO)est (i )
i =1 i =1
In order to demonstrate the approach, the errors for the second data set are illustrated
in Figure 7-9 and the average absolute error sums are given in Table 7-2.
74/188
The error sums are not exactly zero, which implies that the steady state assumption is
not fulfilled in the constant parameter approach. However, considering the possible
inaccuracy of the offline measurements, the fit is sufficient. Thus, the constant
efficiency approach is tested in the simulation study.
Table 7-2: Average absolute errors for the different copper concentrations.
Figure 7-9: The scaled values of the loaded organic, rich electrolyte, partial PLS and barren
organic copper concentrations; measured (red, solid) and estimated (blue, dotted) for the second
data set.
7.2.2.1 Efficiency and recycle correction parameters for the first test
period
The constant efficiency and recycle correction parameters are optimized for the first
offline test data set. The parameter values are shown in Table 7-3.
Table 7-3: Efficiencies and recycle corrections for nominal (NOM) and changing equilibrium
isotherm (I) models.
Model α EP α E1 α E2 αS cfEP cf E1 cf E2 cf S
(NOM) 0.9591 1.0036 0.8892 0.8351 0.9728 1.1951 0.9854 0.9523
(I) 0.9710 1.0670 0.7900 0.8432 0.9787 1.0720 0.8836 0.9171
The varied efficiency and recycle correction parameters are calculated from the first
offline data set. The parameter values for the first test period are shown in Figure 7-10
and Figure 7-11. All the parameters are close to 1, although larger variation occurs
75/188
simultaneously with very high or low reagent volume per cent points. This is natural
since the efficiency is strongly dependent on the equilibrium isotherm. The
parameters for stripping do not change significantly, since the equilibrium isotherm
parameters are relatively stable, as can be seen from Figure 7-10.
Figure 7-10: Varied efficiencies for the extraction and stripping stages.
Figure 7-11: Estimated recycles for the extraction and stripping stages.
76/188
Table 7-4: Efficiencies and recycle corrections for nominal (nom) and changing equilibrium
isotherm (I) models.
Model α EP α E1 α E2 αS cfEP cf E1 cf E2 cf S
(NOM) 0.9421 0.9478 0.9577 0.8783 1.0738 1.2738 1.0388 1.1299
(I) 1.0859 0.6655 1.0157 0.9242 1.0519 1.3181 0.9401 1.0195
The varied efficiency and recycle correction parameters are calculated for the second
test data set. The parameter values are shown in Figure 7-12 and Figure 7-13. The
efficiencies are slightly lower than for the first test data set, and there is much more
variation in the recycle correction parameters, due to sudden drops in the flow rates
and concentrations.
Figure 7-12: Varied efficiencies for the extraction and stripping stages.
77/188
Figure 7-13: Estimated recycles for the extraction and stripping stages.
The time delays of the plug flow models for the settlers and the organic storage tank
are calculated on the basis of the volumes and average flow rates. The time delays of
the first and second test periods are presented in Table 7-5.
Table 7-5: Time delays of the plug flow model parts for the first test period. The unit is one
sampling time.
5 ei , j
total j = ∑ (7-7)
i =1 ei , NOM
Since there are five outputs, the total error sum for the nominal model is 5.
7.3.2 Simulation performance for the first test period
The input data and the parameters of the first test period are described in Chapters 7.1
and 7.2.
The statistical results for the first test period are presented in Table 7-6 and Table 7-7.
In the tables, the columns represent the different model structures. The first five rows
present the error sums for each output copper concentration, and the sixth row the
total sum for each column.
The results clearly show that adaptation of the equilibrium isotherm parameters
improves the result. Adding the efficiency parameter adaptation decreases the
residuals even more for all the other variables, except for the loaded organic copper
concentration. Adding the mass transfer parameter does not improve the result and,
therefore, the visual model comparison is performed only for the nominal (NOM),
equilibrium isotherm (I), and equilibrium isotherm with efficiency (EI) varied models.
Table 7-6: Average absolute error percentage for the different models.
Table 7-7: Average integral of absolute error for the different models.
NOM I EI EKI
c(LO) 0.31 0.28 0.32 0.33
c(BO) 0.18 0.22 0.13 0.14
c(RE) 0.64 0.60 0.57 0.57
c(RaffS) 0.11 0.05 0.04 0.04
c(RaffP) 0.08 0.03 0.02 0.02
total 5 3.89 3.26 3.35
For the rich electrolyte copper concentration the changes are well predicted with all
the models, although at the beginning and end of the test period there is a clear
difference. The good fit to the model can be explained by the dependence on the lean
electrolyte concentration, which causes a similar effect in all the models. The
deviation at the end is caused by analyzer calibration around sampling time 2800.
80/188
Figure 7-14: Rich electrolyte copper concentration; measured (dotted, black), nominal (nom)
model (dashed, blue), equilibrium isotherm parameter varied (I) model (solid, red), and
equilibrium isotherm and efficiency parameter varied (EI) model (dash dotted, magenta).
The loaded organic copper concentration was extremely noisy before filtering, and the
reliability of the filtered measurements was not very good either. Therefore, for this
test period the trends are more important. The model with adaptation in only the
equilibrium isotherm parameters best fits the industrial data. However, the organic
copper concentrations generally should have similar trends, and here the barren
organic copper concentration measurement was more reliable. Therefore, the model
with both parameters varied might be the best one, since it fits best to the trends in the
barren organic copper concentrations, as presented in Figure 7-16. The rise in the
level at 2800 due to analyzer calibration cannot be captured by any of the models.
Figure 7-15: Loaded organic copper concentration; measured (dotted, black), nominal (nom)
model (dashed, blue), equilibrium isotherm parameter varied (I) model (solid, red), and
equilibrium isotherm and efficiency parameter varied (EI) model (dash dotted, magenta).
The fit of the EI model to the barren organic data trends is relatively good, whereas
the nominal model has the worst fit. The model with adaptation in only the
equilibrium isotherm model (I) is also slightly off the trends. The efficiency and
recycle correction parameter adaptation clearly helps to capture this un-modeled
phenomenon.
81/188
Figure 7-16: Barren organic copper concentration; measured (dotted, black), nominal (nom)
model (dashed, blue), equilibrium isotherm parameter varied (I) model (solid, red), and
equilibrium isotherm and efficiency parameter varied (EI) model (dash dotted, magenta).
Figure 7-17: Raffinate series copper concentration; measured (dotted, black), nominal (nom)
model (dashed, blue), equilibrium isotherm parameter varied (I) model (solid, red), and
equilibrium isotherm and efficiency parameter varied (EI) model (dash dotted, magenta).
For the raffinate parallel copper concentration the nominal model has different level
than the measurements, but the trends are similar. The level is corrected by
introducing adaptation to the equilibrium isotherm parameters, and the fit is further
increased with the efficiency parameter adaptation.
82/188
Figure 7-18: Raffinate parallel copper concentration; measured (dotted, black), nominal (nom)
model (dashed, blue), equilibrium isotherm parameter varied (I) model (solid, red), and
equilibrium isotherm and efficiency parameter varied (EI) model (dash dotted, magenta).
The model with adaptation in both the equilibrium isotherm and efficiency parameters
described the industrial data the best. The model fit to the industrial data was
adequate. The model follows the main trends caused by flow rate, concentration and
chemical changes. In order to get more confidence in the model, the second data set
was tested with the same model structures.
7.3.3 Simulation performance for the second test period
The input data and the parameters of the second test period are described in Chapters
7.1 and 7.2.
The statistical results for the second test period are presented in Table 7-8 and
Table 7-9. In the tables, the columns represent the different model structures. The first
five rows present the error sums for each output copper concentration, and the sixth
row the total sum for each column.
The results clearly imply that including adaptation in both the equilibrium isotherm
and efficiency parameters is necessary. Adaptation of the mass transfer coefficient
does not improve the predictions and is therefore not necessary. With the EI model
the average absolute error percentages are at a satisfactory level for the electrolyte and
organic copper concentrations and, considering the noise level of the raffinate copper
concentrations, the raffinate predictions are very good.
Table 7-8: Average absolute error percentage for the different models.
Table 7-9: Average integral of absolute error for the different models.
Figure 7-19: Rich electrolyte copper concentration; measured (dotted, black), nominal (nom)
model (dashed, blue), equilibrium isotherm parameter varied (I) model (solid, red), and
equilibrium isotherm and efficiency parameter varied (EI) model (dash dotted, magenta).
Figure 7-20 presents the process data and model predictions for the loaded organic
copper concentration. The constant parameter approach to the model loaded organic
copper concentration is not successful because the change in the organic copper level
is due to the increase in the reagent volume per cent in the organic solution. The
varied equilibrium isotherm model (I) successfully predicts the change, but the level
is not correct due to the constant efficiencies of the extraction stages. The fit of the
equilibrium isotherm and efficiency parameter model (EI) predictions match well the
loaded organic copper concentration measurements. This experiment clearly
demonstrates the need to change both the equilibrium isotherm and efficiency
parameters simultaneously.
84/188
Figure 7-20: Loaded organic copper concentration; measured (dotted, black), nominal (nom)
model (dashed, blue), equilibrium isotherm parameter varied (I) model (solid, red), and
equilibrium isotherm and efficiency parameter varied (EI) model (dash dotted, magenta).
The barren organic copper concentration measurements and model predictions are
presented in Figure 7-21. The quality of the barren organic copper concentration
predictions are very similar to that of the loaded organic copper concentrations. The
constant parameter model (NOM) cannot adapt to the change in the reagent volume
per cent. The equilibrium isotherm model (I) sufficiently well predicts the changes,
and the (EI) model even better. Adaptation in the equilibrium isotherm and efficiency
parameters is clearly necessary.
Figure 7-21: Barren organic copper concentration; measured (dotted, black), nominal (nom)
model (dashed, blue), equilibrium isotherm parameter varied (I) model (solid, red), and
equilibrium isotherm and efficiency parameter varied (EI) model (dash dotted, magenta).
Figure 7-22 presents the process data and the model predictions for the raffinate series
copper concentration. The raffinate series predictions suffer drastically from the
deviations between the true and utilized equilibrium isotherm and efficiency
parameters. The constant extraction isotherm parameters give too high an extraction
85/188
rate, which pushes the raffinate to the minimum allowed level, as can be seen for the
nominal (NOM) and equilibrium isotherm model (I) experiments. The predictions
with varied efficiency and equilibrium isotherm parameters are very reasonable and
match well with the measurement data.
Figure 7-22: Raffinate series copper concentration; measured (dotted, black), nominal (nom)
model (dashed, blue), equilibrium isotherm parameter varied (I) model (solid, red), and
equilibrium isotherm and efficiency parameter varied (EI) model (dash dotted, magenta).
Figure 7-23 presents the process data and the model predictions for the raffinate
parallel copper concentration. The quality of the raffinate parallel copper
concentration predictions is similar to the quality of the raffinate series copper
concentration predictions. Due to the change in the reagent level in the organic
solution, nominal model prediction fails. For the equilibrium isotherm varied model
(I) the efficiency is too high, and the raffinate level is almost at the minimum value
possible. This is therefore corrected by adding adaptive efficiency. The match
between the measurements and the varied (EI) model are relatively good.
86/188
Figure 7-23: Raffinate parallel copper concentration; measured (dotted, black), nominal (nom)
model (dashed, blue), equilibrium isotherm parameter varied (I) model (solid, red), and
equilibrium isotherm and efficiency parameter varied (EI) model (dash dotted, magenta).
The interplay between the equilibrium isotherm and the efficiency is demonstrated by
the organic and raffinate copper concentrations: the changes in the reagent volume per
cent induce a change in the equilibrium. The equilibrium isotherm adaptation is thus
necessary and, on the basis of the experiments, adaptation of the efficiencies is
crucial. The average absolute error percentages are reasonably good for all the
measurements with the (EI) model: less than 2% for the electrolyte and organic
copper concentrations, and less than 10% for the less reliable raffinate copper
concentrations.
7.3.4 Simulation performance under input changes
The simulation performance is further evaluated with five examples with distinct
changes in the input variables. The examples are chosen from the two test data sets.
The simulation performance is evaluated with input flow rate changes in Section
7.3.4.1, then with a change in the input copper concentrations in Section 7.3.4.2, and,
finally, with a change in the reagent volume per cent in the organic solution in Section
7.3.4.3.
Figure 7-24: Rich electrolyte copper concentration for sampling period [800, 1100] in the first
data set; measured (dotted, black), nominal (nom) model (dashed, blue), equilibrium isotherm
parameter varied (I) model (solid, red), and equilibrium isotherm and efficiency parameter
varied (EI) model (dash dotted, magenta).
Figure 7-25: Raffinate parallel copper concentration for sampling period [800, 1100] in the first
data set; measured (dotted, black), nominal (nom) model (dashed, blue), equilibrium isotherm
parameter varied (I) model (solid, red), and equilibrium isotherm and efficiency parameter
varied (EI) model (dash dotted, magenta).
During sampling period [1800, 2100] in the first data set there is a downward peak in
the PLS and organic flow rates, and during sampling period [2000, 2300] there is a
downward peak in the electrolyte flow. These changes cause peaks in the rich
electrolyte and raffinate parallel copper concentrations, as illustrated in Figure 7-26
and Figure 7-27.
For the rich electrolyte copper concentration the downward step in the electrolyte
flow rate causes a downward peak around sampling time 2100. The upward steps in
the organic flow rate and the electrolyte flow rate cause oscillations between [2100 -
2350] in the rich electrolyte copper concentration, and a larger downward peak
around 2300. All the models follow the changes well, the model with adaptation in the
isotherm and efficiency parameters (EI) having the best fit.
88/188
The flow rate changes in the PLS parallel and series flow rates and the organic flow
rate cause a mild upward peak in the raffinate parallel copper concentration around
sampling period [2000, 2100]. The models adequately follow the trends.
Figure 7-26: Rich electrolyte copper concentration for sampling period [2000 - 2300] in the first
data set; measured (dotted, black), nominal (nom) model (dashed, blue), equilibrium isotherm
parameter varied (I) model (solid, red), and equilibrium isotherm and efficiency parameter
varied (EI) model (dash dotted, magenta).
Figure 7-27: Raffinate parallel copper concentration for sampling period [2000 - 2300] in the first
data set; measured (dotted, black), nominal (nom) model (dashed, blue), equilibrium isotherm
parameter varied (I) model (solid, red), and equilibrium isotherm and efficiency parameter
varied (EI) model (dash dotted, magenta).
A drop in the flow rates around sampling period [2200, 2500] in the second data set
causes a downward step in the rich electrolyte copper concentration and an upward
peak in the raffinate series copper concentration, as illustrated in Figure 7-28 and
Figure 7-29. The rich electrolyte copper concentration trends are the best followed by
the model structure with adaptation in both the equilibrium isotherm and efficiency
(EI), although the other models follow the flow rate changes adequately well, too.
The upward peak in the raffinate series concentration between sampling period [2300
- 2450] is best followed by the model structure with adaptation in both the equilibrium
isotherm and efficiency (EI). The nominal model structure has similar trends to the
89/188
measurement data, but the level is well above the raffinate series copper concentration
measurements. The model structure with adaptation in the equilibrium isotherm
parameters (I) dampens the upward peak.
Figure 7-28: Rich electrolyte copper concentration for sampling period [2100, 2600] in the second
data set; measured (dotted, black), nominal (nom) model (dashed, blue), equilibrium isotherm
parameter varied (I) model (solid, red), and equilibrium isotherm and efficiency parameter
varied (EI) model (dash dotted, magenta).
Figure 7-29: Raffinate series copper concentration for sampling period [2100, 2600] in the second
data set; measured (dotted, black), nominal (nom) model (dashed, blue), equilibrium isotherm
parameter varied (I) model (solid, red), and equilibrium isotherm and efficiency parameter
varied (EI) model (dash dotted, magenta).
sampling times) by all the model structures. The lean electrolyte copper concentration
change does not have any significant effect on the raffinate parallel copper
concentration, and the copper concentration remains at the same level with only small
variations. The best model structures are those with varied parameters of the
equilibrium isotherm (I) and the equilibrium isotherm with efficiency (EI). The
downward peak around sampling time 1350 is not explained by any of the models.
Figure 7-30: Rich electrolyte copper concentration for sampling period [1200, 1700] in the first
data set; measured (dotted, black), nominal (nom) model (dashed, blue), equilibrium isotherm
parameter varied (I) model (solid, red), and equilibrium isotherm and efficiency parameter
varied (EI) model (dash dotted, magenta).
Figure 7-31: Raffinate parallel copper concentration for sampling period [800, 1100] in the first
data set; measured (dotted, black), nominal (nom) model (dashed, blue), equilibrium isotherm
parameter varied (I) model (solid, red), and equilibrium isotherm and efficiency parameter
varied (EI) model (dash dotted, magenta).
model structures with adaptation in the equilibrium parameters (I) and adaptation in
both the equilibrium isotherm and efficiency (EI) follow the rising trend in the loaded
organic copper concentration, but are lagged by about 50 sampling times from the
beginning of the change. This is due to the delay in the offline measurement of the
reagent volume per cent in the organic solution. The best model structure is the one
with adaptation in both the equilibrium isotherm and efficiency (EI).
For the rich electrolyte copper concentration the rise in the reagent volume per cent in
the organic solution does not have as drastic an effect as for the loaded organic copper
concentration. This is due to the smaller changes in the linear stripping equilibrium
isotherm. All the model structures follow well the rich electrolyte copper
concentration trends.
92/188
Figure 7-32: Loaded organic copper concentration for sampling period [1100, 1900] in the second
data set; measured (dotted, black), nominal (nom) model (dashed, blue), equilibrium isotherm
parameter varied (I) model (solid, red), and equilibrium isotherm and efficiency parameter
varied (EI) model (dash dotted, magenta).
Figure 7-33: Rich electrolyte copper concentration for sampling period [1100, 1900] in the second
data set; measured (dotted, black), nominal (nom) model (dashed, blue), equilibrium isotherm
parameter varied (I) model (solid, red), and equilibrium isotherm and efficiency parameter
varied (EI) model (dash dotted, magenta).
the simulation performance was gained by using adaptation in the efficiency and
recycle correction parameters. This underlines the necessity of adaptation, especially
for the extraction process in the equilibrium isotherm, efficiency and recycle
correction parameters. Adaptation of the mass transfer parameter did not significantly
affect the results, and thus the constant mass transfer parameters are used in the
further studies.
The smaller peaks in the measurement data were not explained by any of the models.
This might be due to measurement noise or to modeling inaccuracies. The modeling is
unable to describe rapid changes in the reagent volume per cent in the organic
solution and the pH level, because both of these are measured only offline. Adaptation
to the changes in the reagent volume per cent in the organic solution and pH level is
also lagged due to the offline measurement delay.
The model with adaptation in both equilibrium isotherm parameters and efficiency
parameters (EI) was chosen for further studies due to better overall results and a
model structure that gives more information about the process state. The efficiency
parameters describe the unit process efficiencies that seldom are 100% in industrial
plants. Thus, these parameters could be used to give indication of the performance of
the plant.
Variations not captured by this (EI) model can be due to inaccuracy of the data and
process upsets that are can not measured (formation of crud/emulsion, heavy rain).
Therefore, the operating conditions of the process should be verified before applying
the model.
The model can be applied to similar copper solvent extraction plants using mixer-
settlers by modifying the flow configuration between the mixer-settlers, and by
adapting the equilibrium isotherm, efficiency and recycle correction parameters.
94/188
y (t ) − y (0)
G (t ) = (8-1)
u (1) − u (0)
The nonlinearity and asymmetry are evaluated by calculating the percentual absolute
difference between the +5% steady state gain and the steady state gains with different
input change magnitudes. The scaling invariance applies adequately if the
nonlinearity and asymmetry of the responses are mild.
The scaling invariance is tested at the chosen operating points DP1 and DP4 by
introducing ±1%, ±5%, and ±10% changes to all the seven input variables of the
dynamic process models and then collecting the responses of the output variables, rich
electrolyte and loaded organic copper concentrations. A maximum change of ±10% is
tested in order to get more confidence about the results.
The steady state gains, i.e. the gains after the maximum response is reached, are first
determined. The steady state gains for each input-output variable pair are compared in
order to determine whether there are any nonlinearities or asymmetries. The responses
for each input-output variable pair are then plotted in order to confirm the results for
the whole time range.
Table 8-1: Steady state gains for rich electrolyte copper concentration responses at operating
point DP1.
The responses of the rich electrolyte copper concentration to the step input changes at
the first operating point DP1 are presented in Figure 8-1 and Figure 8-2. The
responses appear to be linear with first order plus time delay dynamics, except for the
F(LO) response, which is of the second order with zero plus time delay type. The PLS
and electrolyte copper concentrations, c(PLS) and c(LE), together with the electrolyte
flow rate, F(LE), have the largest impact on the rich electrolyte copper concentration.
96/188
Figure 8-1: Responses of the rich electrolyte copper concentration to input changes in F(PLSS),
F(PLSP), F(LO),F(LE), c(PLS),and c(LE) at operating point DP1.
The response to reagent volume per cent change is especially interesting with inverse
dynamics, as can be seen from Figure 8-2. Since the exact time dynamics of reagent
volume blending to the organic solution are not known, blending is assumed to be
instant. At this operating point DP1, the effect of an increasing reagent volume per
cent is small for a rich electrolyte copper concentration. The inverse effect is due to
the increase in the organic solution copper concentration via the organic recycle
between the stripping and extraction steps.
Figure 8-2: Responses of the rich electrolyte copper concentration to input changes in the reagent
volume per cent at operating point DP1.
The steady state gains for the rich electrolyte copper concentration at the second
operating point DP4 are shown in Table 8-2. The gains are relatively similar between
the positive and negative input steps, except for the organic flow rate F(LO) and
reagent volume per cent where there are high asymmetries of up to 50%, and mild
nonlinear behaviour of up to 35%. For the change in the PLS copper concentration the
asymmetry and nonlinearity are around 10%.
97/188
Table 8-2: Steady state gains for rich electrolyte copper concentration responses at operating
point DP4.
The responses of the rich electrolyte copper concentration to the input step changes at
operating point DP4 are presented in Figure 8-3 and Figure 8-4. The responses mainly
follow the first order plus time delay dynamics, except for the organic flow rate,
F(LO), which is of the second order with zero plus time delay type. The PLS and
electrolyte copper concentrations, c(PLS) and c(LE), and the electrolyte flow rate,
F(LE), have the greatest effect on the rich electrolyte copper concentration.
Figure 8-3: Responses of the rich electrolyte copper concentration to input changes in F(PLSS),
F(PLSP), F(LO),F(LE), c(PLS),and c(LE) at operating point DP4.
The response to a change in the reagent volume per cent is especially interesting with
inverse dynamics, as can be seen from Figure 8-4. The effect of increasing the reagent
volume per cent at operating point DP4 is larger than at the first operating point
(DP1). However, decreasing the reagent volume per cent results in decreased
stripping after the increased stripping period, as at the first operating point for the rich
electrolyte copper concentration. The inverse effect is due to the increase in the
organic solution copper concentration via the organic recycle between the stripping
and extraction steps.
98/188
Figure 8-4: Responses of the rich electrolyte copper concentration to input changes in the reagent
volume per cent at operating point DP4.
Table 8-3: Steady state gains for loaded organic copper concentration responses at operating
point DP1.
The responses of the loaded organic copper concentration to the input step changes at
operating point DP1 are presented in Figure 8-5 and Figure 8-6. All the step responses
follow first order plus time delay dynamics. The main affecting inputs are the PLS
copper concentration and organic flow rate, c(PLS) and F(LO), and the reagent
volume percent.
99/188
Figure 8-5: Responses of the loaded organic copper concentration to input changes in F(PLSS),
F(PLSP), F(LO),F(LE), c(PLS),and c(LE) at operating point DP1.
The response of a change in the reagent volume per cent at operating point DP1 has a
larger effect than a change in the organic flow rate, as can be seen from Figure 8-6
and Figure 8-5. The response is relatively linear and symmetric, with dynamics of a
higher order plus time delay form. An increasing reagent volume per cent in the
organic solution increases the organic copper concentration due to the larger copper
ion complexation potential.
Figure 8-6: Responses of the loaded organic copper concentration to input changes in the reagent
volume per cent at operating point DP1.
The steady state gains for the loaded organic copper concentrations at the second
operating point DP4 are presented in Table 8-4. The gains are relatively linear, and
the asymmetries and nonlinearities are weak, between 10% and 20%, for the organic
and electrolyte flow rates, F(LO) and F(LE), and the reagent volume percent.
100/188
Table 8-4: Steady state gains for loaded organic copper concentration responses at operating
point DP4.
The responses of the loaded organic copper concentration to the input step changes at
operating point DP1 are presented in Figure 8-7 and Figure 8-8. As at operating point
DP1, the step responses all seem to be of a first order plus time delay form. The main
affecting inputs are the PLS copper concentration and the organic flow rate, c(PLS)
and F(LO), and the reagent volume percent.
Figure 8-7: Responses of the loaded organic copper concentration to input changes in F(PLSS),
F(PLSP), F(LO),F(LE), c(PLS),and c(LE) at operating point DP4.
The response of the reagent volume per cent change is slightly larger than the change
in the organic flow rate, as can be seen from Figure 8-8 and Figure 8-7. The response
is weakly nonlinear (15%) and asymmetric (20%), with dynamics of higher order plus
time delay. Increasing the reagent volume per cent in the organic solution naturally
increases the organic copper concentration due to the larger copper ion complexation
potential.
101/188
Figure 8-8: Responses of the loaded organic copper concentration to input changes in the reagent
volume per cent at operating point DP4.
8.1.2 Additivity
Additivity is tested by changing two or more inputs at the same time and comparing
the output change to the sum of the output changes from experiments in which one
input is changed at a time. For example, for the loaded organic copper concentration,
simultaneous changes in the PLS series flow rate and the PLS copper concentration
(combo) are compared to the sum of the loaded organic copper concentration changes
(sum) for the separate experiments in which the PLS series flow rate and the PLS
copper concentration are changed. This is denoted as:
Additivity of the responses of the rich electrolyte and loaded organic copper
concentration is studied at operating point DP4 with 5% step changes to the input
variables. Since the number of all possible input combinations is relatively high, only
the most common cases, with changes in two inputs at the same time are studied. The
collected outputs are compared to the sum of outputs from separate 5% input step
change experiments. The tested input combinations are:
• both PLS series and parallel flow rates, F(PLSS) and F(PLSP)
• PLS series flow rate and organic flow rate, F(PLSS) and F(LO)
• organic and electrolyte flow rate, F(LO) and F(LE)
• PLS series flow rate and PLS copper concentration, F(PLSS) and c(PLS)
• organic flow rate and electrolyte copper concentration, F(LO) and c(LE)
• electrolyte flow rate and electrolyte copper concentration, F(LE) and c(LE)
• organic flow rate and PLS copper concentration, F(LO) and c(PLS)
102/188
Table 8-5: Differencies between the gains of the outputs of combined input changes and sum of
separate input changes for rich electrolyte copper concentration responses in operating point
DP4.
Table 8-6: Percentual differencies between the gains of the outputs of combined input changes
and sum of separate input changes for rich electrolyte copper concentration responses in
operating point DP4.
Figure 8-9: Responses of the rich electrolyte copper concentration to 5% changes in the series
and parallel PLS flow rates, F(PLSS) and F(PLSP). Response from simultaneous input changes
in solid curves (combo), and the sum of responses from separate experiments with dashed curves
(sum).
Table 8-7: Differencies between the gains of the outputs of combined input changes and sum of
separate input changes for loaded organic copper concentration responses in operating point
DP4.
Table 8-8: Percentual differencies between the gains of the outputs of combined input changes
and sum of separate input changes for loaded organic copper concentration responses in
operating point DP4.
Figure 8-10: Responses of the loaded organic copper concentration to 5% changes in the PLS
series and organic flow rates, F(PLSS) and F(LO). Response from simultaneous input changes in
solid curves (combo), and the sum of responses from separate experiments with dashed curves
(sum).
plus time delay form, except for the organic flow rate response to the rich electrolyte
copper concentration and all the reagent volume per cent responses, which are of the
second order with zero plus time delay form.
h11 ⋅ e − h13s org h21 ⋅ e − h23s aq h31 ⋅ e − h33s org h41 ⋅ e − h43s aq
corg
( s) = cin ( s ) + cin ( s ) + Fin ( s ) + Fin ( s )
h12 s + 1 h22 s + 1 h32 s + 1 h42 s + 1
out
(8-3)
Assume that changes in the input concentration cause responses with identical time
dynamics, but different gains. The responses for the input flow rate changes are also
assumed to behave similarly, but with different gains. This yields:
h13
1− s
2 2 − h13 s
e− h13 s ≈ = (8-5)
h 2 + h13 s
1 + 13 s
2
Now the output copper concentrations can be formulated as follows:
107/188
2 − h13 s 1 2 − h33 s 1
org
cout ( s) ≈ h11 ⋅ cinorg ( s ) + h21 ⋅ Cinaq ( s ) + h31 ⋅ Finorg ( s ) + h41 ⋅ Finaq ( s )
2 + h13 s h12 s + 1 2 + h33 s h32 s + 1
(8-6)
org
cout ( s ) = h1 ( s )cinorg ( s ) + h2 ( s )cinaq ( s ) + h3 ( s ) Finorg ( s ) + h4 ( s ) Finaq ( s ) (8-7)
As the volume of the tank is significantly smaller than that of one settler, in the
following analysis the organic tank and the organic phase of the wash stage are
approximated with a constant organic time delay (θ).
Since the output flow rates from settlers S1H, E1P, E1S and E2S, marked as F(BO),
F(BO1), F(BO2), F(PLS1), are not measured, they have to be approximated from the
flow rates and equipment volumes. In the process the mixer and settler volumes are
approximately constant, and the maximum deviation from the normal operating point
during process changes is about 5%. The mixers and settlers are identical and the
organic surface depths are very similar in each settler. Assume that the flow rate is
approximately constant and the settler aqueous and organic volumes are not changing,
then the time delays in the mixer-settlers for the organic (θ) and aqueous (δ) phases
are approximately constant. Now the missing flow rate measurements can be
expressed as:
F ( BO)( s ) = F ( LO) ⋅ e−θ s (8-8)
The detailed models for each of the unit process are first developed, and these models
are then combined to model the output copper concentrations of rich electrolyte and
loaded organic as functions of the measured input variables.
108/188
The output copper concentrations of the first extraction unit process E1P are
formulated as follows:
The output copper concentrations of the second extraction unit process E1S are
formulated as follows:
The output copper concentrations of the third extraction unit process E2S are
formulated as follows:
The output copper concentrations of the stripping unit process S1H are formulated as
follows:
Now the output concentrations of loaded organic c(LO) and rich electrolyte c(RE) can
be expressed as a combination of Equations 8-15 – 8-26 as follows:
1
c( LO)( s ) = ⋅
1 − f 2 ( s )k1 ( s ) − g1 ( s) f1 ( s )h1 ( s ) j1 ( s )
c( PLS )( s ) [ g1 ( s ) f 2 ( s )k2 ( s ) + g1 ( s ) f1 ( s )h2 ( s ) + (1 − f 2 ( s )k1 ( s )) g 2 ( s )]
+c ( LE )( s ) [ g1 ( s ) f1 ( s )h1 ( s ) j2 ( s ) ]
+ F ( LO)( s ) g1 ( s ) f 2 ( s ) k3 ( s )e −3θ s + g1 ( s ) f 3 ( s )e −2θ s + g1 ( s ) f1 ( s )h3 ( s )e −θ s (8-20)
+ g1 ( s ) f1 ( s ) h1 ( s ) j3 ( s ) + (1 − f 2 ( s ) k1 ( s )) g3 ( s )e −3θ s
+ F ( PLSS )( s) g1 ( s ) f 2 ( s )k4 ( s ) + g1 ( s ) f 4 ( s )e −δ s + (1 − f 2 ( s )k1 ( s )) g 4 ( s )
+ F ( PLSP)( s ) [ g1 ( s ) f1 ( s )h4 ( s ) ]
+ F ( LE )( s ) [ g1 ( s ) f1 ( s )h1 ( s ) j4 ( s ) ]
p1 ( s )
c( RE )( s ) = ⋅
1 − f 2 ( s )k1 ( s ) − g1 ( s ) f1 ( s )h1 ( s ) j1 ( s )
c( PLS )( s ) [ g1 ( s ) f 2 ( s )k2 ( s ) + g1 ( s ) f1 ( s )h2 ( s ) + (1 − f 2 ( s )k1 ( s )) g 2 ( s )]
+c ( LE )( s ) [ g1 ( s ) f1 ( s )h1 ( s ) j2 ( s ) ]
+ F ( LO)( s ) g1 ( s ) f 2 ( s ) k3 ( s )e −3θ s + g1 ( s ) f 3 ( s )e −2θ s + g1 ( s ) f1 ( s )h3 ( s )e −θ s
(8-21)
+ g1 ( s ) f1 ( s ) h1 ( s ) j3 ( s ) + (1 − f 2 ( s ) k1 ( s )) g3 ( s )e −3θ s
+ F ( PLSS )( s) g1 ( s ) f 2 ( s )k4 ( s ) + g1 ( s ) f 4 ( s )e −δ s + (1 − f 2 ( s )k1 ( s )) g 4 ( s )
+ F ( PLSP)( s ) [ g1 ( s ) f1 ( s )h4 ( s ) ]
+ F ( LE )( s ) [ g1 ( s ) f1 ( s )h1 ( s ) j4 ( s ) ]
+ p2 ( s )c( LE )( s ) + p3 ( s ) F ( LO)( s ) + p4 ( s ) F ( LE )( s )
Using Equations 8-11 – 8-13, the organic recycle term, consisting of the term
f2(s)k2(s), is caused by the PLS recycle in the series extraction units and term
g1(s)f1(s)h1(s)j1(s) by the loaded organic recycle through all the unit processes, can be
analyzed, as follows:
1
1 − f 2 ( s )k1 ( s ) − g1 ( s ) f1 ( s )h1 ( s ) j1 ( s )
1
= − f 23 s − k23 s
f 21 ⋅ e k ⋅e g ⋅ e − g13s f11 ⋅ e − f13s h11 ⋅ e − h13s j11 ⋅ e − j13s
1− ⋅ 21 − 11
f 22 s + 1 k22 s + 1 g12 s + 1 f12 s + 1 h12 s + 1 j12 s + 1
1
=
1 2 − f13 s f 21 ⋅ k21 2 − k23 s g11 ⋅ f11 h j 2 − ( g13 + h13 + j13 ) s
1− ⋅ ⋅ − ⋅ 11 ⋅ 11 ⋅
f12 s + 1 2 + f13 s k22 s + 1 2 + k23 s g12 s + 1 h12 s + 1 j12 s + 1 2 + ( g13 + h13 + j13 ) s
110/188
( f12 s + 1)( 2 + f13 s1)( k22 s + 1)( 2 + k23 s )( g12 s + 1)( h12 s + 1)( j12 s + 1) ( 2 + ( g13 + h13 + j13 ) s )
=
( f12 s + 1)( 2 + f13 s1)( k22 s + 1)( 2 + k23 s )( g12 s + 1)( h12 s + 1)( j12 s + 1) ( 2 + ( g13 + h13 + j13 ) s )
+ ( 2 − f13 s ) ⋅ f 21 k21 ( 2 − k23 s )( g12 s + 1)( h12 s + 1)( j12 s + 1) ( 2 + ( g13 + h13 + j13 ) s )
− ( 2 − f13 s ) ⋅ g11 f11 h11 j11 ( k22 s + 1) ( 2 + k23 s ) ( 2 − ( g13 + h13 + j13 ) s )
(8-22)
The order of the numerator is 8 and the orders of the denominator terms are 8, 6 and
4. If the organic recycle term can be neglected, from Equations (8-27) and (8-28),
then the highest denominator orders for the loaded organic copper concentration are
[6 8 8 6 6 8] and for the rich electrolyte concentration [8 10 10 8 8 10], with the
corresponding inputs [c(PLS), c(LE), F(PLSS), F(PLSP), F(LO), F(LE)]. If the
recycle (Equation 8-29) of the highest denominator order of 8 is included, then the
highest denominator order is 16 for loaded organic and 18 for rich electrolyte. Thus,
for the state-space model identification, model orders of between 6 and 18 should be
tested.
(8-23)
, where the number (n) of states x determines the order of the coefficient matrices
M 1 ∈ R n×n , M 2 ∈ R n×7 , M 3 ∈ R 2×n .
Matrix M4 in this case is [0] since there are no direct effects from the inputs to the
outputs. (Ljung, 1987, pp. 82-86)
The transfer function model (TF) parameters are identified from the PRBS data
separately for both outputs. The transfer function model forms are predetermined on
the basis of the step responses (Section 8.1.1). Most of the submodels are of the first
order plus time delay form. The more complex dynamics are modeled as second order
with zero plus time delay. The transfer function models for loaded organic and rich
electrolyte copper concentrations, c(LO) and c(RE), are of the following form:
percentage, the better the fit. The fit index is especially suitable when the comparison
is performed at one operating point. (Matlab system identification toolbox: Ljung,
2006).
where Ymeas is the measured value, Ymeas is the average value of the measured outputs,
Ymodel is the model value. The fit index values are between [-100 100].
Table 8-9: Model fits to the rich electrolyte copper concentration responses to the 5% input steps
(Valid 1) at DP1 and DP4, to mechanistic model outputs (Valid2) with the inputs of the first and
second industrial data set, and to the first and second industrial data set (Valid3).
Figure 8-11: Rich electrolyte copper concentration; measurement of the first industrial data set
(solid), mechanistic model (dashed), transfer function model (dotted), and eight order state space
model (dash dotted).
The rich electrolyte copper concentration measurements of the second industrial data
set (Valid3 DP4) are well followed by the linearized models, as can be seen from
Figure 8-12. The dynamics are very similar to the process and the mechanistic model
data, with a slight difference in the end of the data set.
114/188
Figure 8-12: Rich electrolyte copper concentration; measurement of the second industrial data
set (solid), mechanistic model (dashed), transfer function model (dotted), and eight order state
space model (dash dotted).
Table 8-10: Model fits to the loaded organic copper concentration responses to the 5% input
steps (Valid1) at DP1 and DP4, mechanistic model outputs (Valid2) with the inputs of the first
and second industrial data set, and to the first and second industrial data set (Valid 3).
At the first operating point DP1, the linearized models follow the trends of the
dynamic process models (Valid2 DP1) with a slight difference between the levels.
The process data (Valid3 DP1) for the loaded organic copper concentration are of
poor quality, but the state space model was relatively successful in following the
trends. The eight order state space model is closer to the mechanistic model, as can be
seen from Figure 8-13.
Figure 8-13: Loaded organic copper concentration; measurement (solid), mechanistic model
(dashed), transfer function model (dotted), and eight order state space model (dash dotted).
At the second operating point DP4, the transfer function outputs and the outputs of the
state space model with eight states are plotted against the second industrial data set
(Valid3 DP4) with reasonably good fit, as can be seen in Figure 8-14. The value of the
loaded organic copper concentration changes much faster with the dynamic process
models than with the linear models, but the final level is reached relatively
successfully with all the models. The linearized models have the same dynamic
changes as the mechanistic models, but they slightly lack the level adaptation for the
loaded organic copper concentration.
116/188
Figure 8-14: Loaded organic copper concentration; measurement (solid), mechanistic model
(dashed), transfer function model (dotted), and eight order state space model (dash dotted).
9.1 Control objectives and the current control strategy of the case
copper solvent extraction plant
Maximization of copper production is the main goal of the control strategy in the
copper solvent extraction and electrowinning plant. The control strategy in the plant
for solvent extraction is to keep the flow rates as high as possible in order to
maximize copper mass flow through the process. The control strategy in the plant for
the electrowinning process is to keep the current amperages as high as economically
possible in order to maximize the copper cathode production.
In the copper solvent extraction process, the regulatory control level of the automation
system consists of PID loops for the flow rates. Higher control levels do not exist. The
operators choose the flow rate setpoints that keep the process within the target values.
The metallurgists give target values for the lean electrolyte copper concentrations
once a week. The flow rates have restrictions due to the maximum pumping
capacities, and the aqueous to organic flow ratios maintaining the phase continuities
in the mixers.
Maximum production is not achieved with the current control strategy. The long time
delays between the control actions and the responses in the key process variables,
complex interactions and cause-effect relationships make control of the process a very
challenging task for a human operator. Therefore the control actions are very
conservative and the process is in a suboptimal state for most of the time. The current
control strategy also lacks real time optimization that would enable steering the
process to the optimal operating point.
118/188
OPTIMIZATION
maximize production
STABILIZING CONTROL
SISO/MIMO strategy
Setpoints for the flow
rates (MV )
BASIC CONTROL
PID loops for the flow rates
INSTRUMENTATION
Flow control valves, flow rate meas,online
analyzers, pH sensor, …
Figure 9-1: Proposed control hierarchy for the copper solvent extraction process. The
optimization layer provides the setpoints of the controlled variables to the supervisory control
level. The stabilizing control level is based on a single-input single-output or multi-input multi-
output control strategy. The stabilizing control level provides the setpoints of the manipulated
variables to the regulatory control level. The basic control level gives signals to the final control
elements at the instrumentation level. The measurement information is led from the
instrumentation level to all the upper levels.
The possible controlled variables (CV) are the outputs of the extraction, the loaded
organic and raffinate copper concentrations, and the outputs of stripping, and the rich
electrolyte and barren organic copper concentrations. Since only rich electrolyte
solution enters the electrowinning process, the rich electrolyte copper concentration is
chosen as the primary controlled variable. The raffinate copper concentration
measurements are unreliable, as stated in Chapter 5.2, and thus the loaded organic
copper concentration is chosen as the secondary controlled variable.
The available manipulated variables (MV) are the flow rates of PLS, organic and
electrolyte, F(PLSS), F(PLSP), F(LO) and FLE).
The measured disturbance variables (DV) are the PLS and lean electrolyte
concentrations and the total PLS flow rate, and the unmeasured disturbance is the
change in reagent volume per cent in the organic solution and pH changes in PLS and
acidity changes in the electrolyte solution.
120/188
The restrictions are the organic to aqueous ratio, which is related to phase continuity,
pumping capacity, and the organic level in the tanks. The process, together with a
classification of the controlled, manipulated and disturbance variables, is presented in
Figure 9-1.
c(LO) F(LO)
Extraction Tank Stripping
Organic solution
c(RaffP)c(RaffS) c(RE)
Figure 9-2: Solvent extraction process. Controlled variables (CV) are marked in bold,
manipulated variables (MV) underlined, and disturbance variables marked in italics.
x& = M1 x + M 2 u
(9-1)
y = M 3 x + M 4u
where x are states, u inputs and y outputs. In this notation, n is the number of states, m
the number of inputs and p the number or outputs. The order of the coefficient
matrices are defined as M 1 ∈ R n×n , M 2 ∈ R n×m , M 3 ∈ R p×n and M 4 ∈ R p×m .
The observability matrix is determined as:
M3
M M
3 1
Obs = M 3 M1 (9-2)
T 2
M
M M n−1
3 1
If the rank (column rank) of the observability matrix is the same as the number of
states n, then the system is state observable. (Ogata, 1987, Skogestad and
Postlethwaite, 2005, p.131)
The state controllability matrix is determined as:
121/188
Co = M 2 M1 M 2 M 12 M 2 L M 1n −1 M 2 (9-3)
If the rank (row rank) of the controllability matrix is the same as the number of states
n, then the system is state controllable. (Ogata, 1987, Skogestad and Postlethwaite,
2005, p.128)
Controllability and observability conditions are directly related to the cancellation of
poles and zeros in the corresponding transfer functions.
In order to study whether the system is input-output controllable, the condition
number of the system has to be low (less than 10), i.e. the directionality of the system
is desired to be weak (Skogestad and Postlethwaite, 2005). The condition number is
calculated as the ratio of the maximum and minimum singular values of the system at
different frequencies.
The stability is studied by plotting the Nyquist plot of all the controlled variable –
manipulated variable transfer function pairs. If the –1 point is not circled, then the
subsystem is stable.
For the transfer function model linearized at the operating point DP1 (DP1TF), the
observability and controllability were studied by comparing the row rank of the
observability matrix and column rank of the controllability matrix to the number of
states of the state space system. Since the ranks and number of states were equal, the
linear model is state observable and state controllable. The condition number,
presented in Figure 9-3, is below 5 for the studied frequency range [0, 0.2] and, thus,
the system is input-output controllable. The stability requirement with a P controller
can also be met; as can be seen from Figure 9-5, all the curves are on the right hand
side of the point -1. (For the Nyquist analysis with the PI controllers, see Section
10.2)
For the transfer function model linearized at the operating point DP4 (DP4TF), the
observability and controllability were studied by comparing the row rank of the
observability matrix and column rank of the controllability matrix to the number of
states of the state space system. Since the ranks and number of states were equal, the
linear model is state observable and state controllable. The condition number,
presented in Figure 9-4 is below 9 over the studied frequency range [0, 0.2] and, thus,
the system is input-output controllable. The stability requirement with a P controller
can also be met; as can be seen from Figure 9-6, all the curves are on the right hand
side of the point -1. (For the Nyquist analysis with the PI controllers, see Section
10.2)
122/188
Figure 9-3: Condition number at different frequencies for the transfer function matrix linearized
at the first operating point DP1.
Figure 9-4: Condition number at different frequencies for the transfer function matrix linearized
at the second operating point DP4.
123/188
Figure 9-5: Nyquist plots of the controlled variable-manipulated variable pairs of the transfer
function model linearized at operating point DP1. The controlled variable on the first row is the
loaded organic copper concentration, c(LO), and on the second row the rich electrolyte copper
concentration, c(RE). The manipulated variables on the columns are: first PLS series flow rate,
F(PLSS), second PLS parallel flow rate, F(PLSP), third organic flow rate, F(LO), and fourth
electrolyte flow rate, F(LE).
Figure 9-6: Nyquist plots of the controlled variable-manipulated variable pairs of the transfer
function model linearized at operating point DP4. The controlled variable on the first row is the
loaded organic copper concentration, c(LO), and on the second row the rich electrolyte copper
concentration, c(RE). The manipulated variables on the columns are: first PLS series flow rate,
F(PLSS), second PLS parallel flow rate, F(PLSP), third organic flow rate, F(LO), and fourth
electrolyte flow rate, F(LE).
Since the linear model of the plant is state controllable, state observable and the
system with P controllers would be stable, the control strategy can be designed and
124/188
tested with the following steps: pairing the controlled and manipulated variables,
designing SISO and MIMO controllers for the stabilizing control level and optimizing
the algorithm for the optimization level, and implementing and testing these
strategies.
( )
T
RGA( M ) = M × M −1 (9-5)
If there are more inputs than outputs, or vice versa, then the gain matrix becomes non-
square, and the inverse operation is replaced by a pseudo-inverse operation.
Skogestad and Postlethwaite (2005) suggest discarding the columns or rows with a
sum of the RGA elements of far less than one. In other cases RGA should be
performed for square sub-matrices.
According to Ogunnaike (1994), the pairing of the input uj with output yi is possible if
the RGAij value is above 0.5 and not too much larger than 1. The optimum value is 1,
which means that there are no interactions from the other inputs with the considered
input. The closer the value is to 0, the less the input affects the output and the bigger
the interactions are with other outputs. The higher the RGAij value, the more the other
loops oppose the effect from the input ui to the output yj. Pairing with negative RGA
values is highly unrecommendable, because the open-loop and closed-loop gains have
opposite signs; the other inputs are more dominant to the output, and also the effect of
the other loops has the opposite effect. To verify the pairing, Glad and Ljung (2004)
suggest evaluation of RGA at other typical frequencies.
In this study, the pairing of the controlled variables (loaded organic and rich
electrolyte copper concentrations, c(LO) and c(RE)), with the manipulated variables
PLS series and parallel, organic and electrolyte flow rates, F(PLSS), F(PLSP), F(LO),
F(LE), was first analyzed using a full non-square matrix RGA for both transfer
function matrices at frequencies ω ∈ [0, 1 ] . Analysis of the first controlled variable,
5
the loaded organic copper concentration c(LO), at the two operating points DP1 and
DP5 are presented in Figure 9-7 and Figure 9-8, respectively. The analysis of the
second controlled variable, the rich electrolyte copper concentration c(RE), at the two
operating points DP1 and DP5 are presented in Figure 9-9 and Figure 9-10.
For the loaded organic copper concentration the pairing at lower frequencies (ω<0.1)
favours the organic flow rate, F(LO), for both cases. For the first case, where the
analysis is performed with the transfer function matrix linearized at the operating
125/188
point DP1, at higher frequencies (ω>0.1) the pairing with the PLS series flow rate,
F(PLSS), becomes more favourable, as illustrated in Figure 9-7. However, the RGA
values for the organic flow rate pairing do not fall below 0, so this pairing is still valid
over the frequency range.
Figure 9-7: The RGA values for the loaded organic copper concentration, c(LO), pairing with the
organic flow rate, F(LO), PLS series flow rate, F(PLSS), PLS parallel flow rate, F(PLSP) and
electrolyte flow rate, F(LE), at frequencies ω=0 … 0.2. The analysis is performed with the
transfer function matrix linearized at the operating point DP1.
The pairing favours the organic flow rate, F(LO), in the second case, where the
analysis is performed with the transfer function matrix linearized at the operating
point DP4, as illustrated in Figure 9-8. The RGA values for this pairing are above 0.6
for the whole frequency range.
126/188
Figure 9-8: The RGA values for the loaded organic copper concentration, c(LO), pairing with the
organic flow rate, F(LO), PLS series flow rate, F(PLSS), PLS parallel flow rate, F(PLSP) and
electrolyte flow rate, F(LE), at frequencies ω=0 … 0.2. The analysis is performed with the
transfer function matrix linearized at the operating point DP4.
For the rich electrolyte copper concentration, the RGA analysis favours pairing with
the electrolyte flow rate. The RGA values for this pairing are above 0.9 at all
frequencies, as illustrated in Figure 9-9 and Figure 9-10.
127/188
Figure 9-9: The RGA values for the rich electrolyte copper concentration, c(RE), pairing with the
organic flow rate, F(LO), PLS series flow rate, F(PLSS), PLS parallel flow rate, F(PLSP) and
electrolyte flow rate, F(LE), at frequencies ω=0 … 0.2. The analysis is performed with the
transfer function matrix linearized at the operating point DP1.
Figure 9-10: The RGA values for the rich electrolyte copper concentration, c(RE), pairing with
the organic flow rate, F(LO), PLS series flow rate, F(PLSS), PLS parallel flow rate, F(PLSP) and
electrolyte flow rate, F(LE), at frequencies ω=0 … 0.2. The analysis is performed with the
transfer function matrix linearized at the operating point DP4.
On the basis of this analysis, the favourable pairing of the controlled and manipulated
variables is the loaded organic copper concentration with the organic flow rate,
c(LO)-F(LO), and the rich electrolyte copper concentration with the electrolyte flow
rate, c(RE)-F(LE).
This pairing is further studied by performing RGA for the [c(LO),c(RE)]x
[F(LO),F(LE)] square matrices at frequencies ω = 0,1/ 20,1/10,1/ 5 . The RGA values
are presented in Table 9-1. The 1-2 pairing values, calculated for the c(LO)-F(LO)
128/188
and c(RE)-F(LE) pairs, are close to one at all the studied frequencies. Therefore, the
pairing between the loaded organic copper concentration with the organic flow rate,
c(LO)-F(LO), and the rich electrolyte copper concentration with the electrolyte flow
rate, c(RE)-F(LE) is chosen for the control strategy development.
Table 9-1: RGA coefficients for 1-2 pairing on the basis of the transfer function matrix linearized
at the operating points DP1 and DP4.
F(PLSS) FT FF1
FB1 FC
F(LE) FC
c(PLS1)
CT
E1P E1S E2S Tank S1H
c(BO1) c(BO2) c(LO) F(LO)
c(BO)
Figure 9-11: Control strategy: the loaded organic copper concentration, c(LO), is feedback
controlled (FB1) with the organic flow F(LO). The rich electrolyte copper concentration, c(RE),
is feedback controlled (FB2) with the electrolyte flow, F(LE). The disturbances are compensated
with feedforward controllers, for PLS series flow, F(PLSS) with FF1, for PLS parallel flow,
F(PLSP), with FF2, for PLS copper concentration, c(PLS), with FF3, and for the lean electrolyte
copper concentration, c(LE) with FF4.
F(PLSS) FF1
SISO strategy
+
F(PLSP) FF2 +
+
c(PLS) FF3
- +
c(LO)sp PI1 F(LO) sp COPPER c(LO)
+ +
SOLVENT
+ + EXTRACTION
c(RE)sp PI2 F(LE) sp c(RE)
- + PROCESS
c(LE) FF4
+
c(LO)sp F(LO) sp COPPER c(LO)
+
MPC SOLVENT
+ EXTRACTION
c(RE)sp F(LE) sp PROCESS c(RE)
+
c(LE) FF4
Figure 9-12: The single-input single-output (SISO) and multi-input multi-output (MIMO) control
strategies. The SISO strategy utilizes two PI controllers marked with PI1 and PI2, and the
MIMO strategy utilizes model predictive controller, marked with MPC. The additional
feedforward compensators are marked with FF. The setpoints are marked with the subscript sp.
Extraction
c(PLS), F(PLSP)
E1P
c(RaffP), F(RaffP)
c(BO1)
F(BO1)
Extraction
E1S
c(PLS1)
F(PLS1) c(RaffS), F(RaffS)
c(BO2)
F(BO2)
+ Stripping
…c(RE) PI2 +
F(LE) S1H c(BO)
F(BO)
c(RE)sp c(LE)
c(RE), F(RE)
…c(LE) FF4
Figure 9-13: Simulation model of the case copper solvent extraction process with two PI
controllers (PI1 and PI2) and four feedforward compensators (FF1, FF2, FF3, FF4). The
manipulated variables are the organic and electrolyte flow rates, F(LO) and F(LE). The inputs to
the PI controllers PI1 and PI2 are the measurements (…c) and the setpoints (sp). In the
feedforward compensators the measurements (...c and …F) are compared to the nominal values
of these variables.
Table 9-2: Manipulated, controlled and disturbance variables, and the parameters of the
optimization problem.
Continuous direct measurement of the production does not exist due to the nature of
the electrowinning process (in electrowinning process copper cathodes are grown for
one week in electrolysis cells, and weighted only after they are taken out of the cells).
, thus the production is estimated from the copper concentrations and flow rates of the
rich and lean electrolyte solutions.
The maximization of production, i.e. the copper mass flow out of the copper solvent
extraction process, can be formulated mathematically as the difference in the copper
concentration between the rich and lean electrolytes, multiplied by the electrolyte
flow rate, as follows:
P = [ c( RE ) − c( LE )] F ( LE ) (9-6)
The restrictions of the optimization problem are the aqueous to organic ratios in the
mixers. In the extraction part the mixers are assumed to run aqueous continuously, i.e.
the major phase is aqueous, and therefore the organic to aqueous ratio has to be below
β1 (<1). In the stripping part, the mixers are assumed to run organic continuously with
an organic to aqueous ratio of above β2 (>1). These restrictions can be formulated for
the extraction as follows:
F ( LO) F ( LO)
≤ β1 ≤ β1 (9-7)
F ( PLSS ) F ( PLSP )
and for the stripping as follows:
133/188
F ( LO)
≥ β2 (9-8)
F ( LE )
The optimization problem can be presented with the controlled, manipulated and
disturbance variables by assuming that the transfer function models of the plant
represent the steady state of the process adequately well, as asserted in Chapter 8.
Now the rich electrolyte copper concentration can be presented as:
c( RE ) = w1 F ( PLSS ) + w2 F ( PLSP ) + w3 F ( LO) − w4 F ( LE ) + w5 c( PLS ) + w6 c( LE ) + w7 vol
(9-9)
and the loaded organic copper concentration as:
c( LO ) = r1 F ( PLSS ) + r2 F ( PLSP) − r3 F ( LO ) − r4 F ( LE ) + r5 c( PLS ) + r6 c( LE ) + r7 vol
(9-10)
The constants, wi and ri, are positive, and represent the absolute values of the steady
state gains of the transfer function models for the loaded organic and rich electrolyte
copper concentrations.
On the basis of these equations (9-9) and (9-10), the profit function (Equation 9-7)
can be expressed using the manipulated and disturbance variables as follows:
w1F ( PLSS ) + w2 F ( PLSP) + w3 F ( LO)
[ c( RE ) − c( LE )] F ( LE ) = + w c( PLS ) − (1 − w )c( LE ) + w vol F ( LE ) − w4 F ( LE )
2
5 6 7
(9-11)
The maximization of this equation requires a maximum value for the organic flow
rate, F(LO). The maximum values of the manipulated flow rates can be derived from
the restrictions of the optimization problem. The organic flow rate has a maximum
restriction in relation to the minimum of the PLS flow rates. Since a maximum value
is desired for the organic flow rate, Equation (9-7) yields:
F ( LO)opt = β1 ⋅ min { F ( PLSS ), F ( PLSP)} (9-12)
The electrolyte flow rate, F(LE), has an optimum point, which can be calculated by
setting the derivative of the profit function (9-11) to zero. The derivative is taken in
relation to the electrolyte flow rate. Using the optimal organic flow rate, F(LO)opt,
Equation (9-12) yields:
−1
F ( LE )opt = w1 F ( PLSS ) + w2 F ( PLSP) + w3 F ( LO) opt + w5 c( PLS ) − (1 − w6 )c( LE ) + w7 vol
2w4
(9-13)
This is the optimal value for the electrolyte flow rate if the maximum electrolyte flow
rate limitation is not exceeded, as required in Equation 9-8. The maximum for the
electrolyte flow rate is smaller than the optimum organic flow rate divided by β2:
1
F ( LE )optF = min F ( LE )opt , F ( LO)opt (9-14)
β2
Now the setpoints for the rich electrolyte and loaded organic copper concentration can
be formulated on the basis of the optimum manipulated variables, the disturbance
variables and optimization parameters, as follows:
134/188
Kp −α p s
G p ( s) = e (10-1)
τ ps +1
1
GC = K C 1 + (10-2)
Ti s
can be derived with the internal model control (IMC) tuning rules. The IMC structure
has only one parameter, λ, to be changed during the tuning procedure. The gain Kc and
the integration time Ti are now determined as follows (Ogunnaike and Ray, 1994, pp.
539):
2τ p + α p
KC = (10-3)
2λ K p
Ti = τ p + 0.5 ⋅ α p (10-4)
The PI controllers were tuned starting from the smallest possible value for λ, and then
increasing the value until adequate performance was reached for both set-point
tracking and disturbance rejection. The coefficients of the PI2 controller in the faster
loop, FB2, were tuned first by keeping the FB1 loop open. The coefficients of the PI1
136/188
controllers in the FB1 loop were then tuned by keeping the FB2 loop open and,
finally, both loops were closed and the coefficients were fine tuned.
The aim of the feedforward controllers is to compensate the effect of the measured
disturbances before they affect the process behaviour. Assuming first order plus time
delay model form for the controlled variable – disturbance variable interactions, then:
K d −α d s
Gd ( s ) = e (10-6)
τ d s +1
The feedforward compensator can be designed by using the CV-DV model and the
CV-MV model as follows (Åström and Wittenmark, 1997, p. 234, Ogunnaike and
Ray, 1994, pp. 571-572):
Gd ( s ) K 1 + τ p s − (α d −α p ) s 1+τ ps
GFF ( s ) = − =− d e ≈ K ff (10-7)
Gp (s) K p 1+τ d s 1+τ d s
The FF controllers were added one by one on the top of the FB controllers, and the
coefficients were tuned.
The SISO control structure with two PI controllers and four feedforward
compensators does not take into account the interactions, and the control actions are
not limited by rate or magnitude.
−α p s
K (T s + 1) K p ⋅ e
GOL = Gc ⋅ G p = c i ⋅ (10-8)
Ti s τ ps +1
The Nyquist stability criterion is used to determine whether the process with PI
controllers is stable (Ogunnaike and Ray, 1994, p.543, Skogestad and Postlethwaite,
2005). The Nyquist plots are illustrated in Figure 10-1 for the process linearized at the
first operating point DP1, and in Figure 10-2 for the process linearized at the second
operating point DP4. Since all the transfer function plots are on the right hand side of
the –1 point, the process with the PI controllers is stable, and there is a gain margin of
about 2.5 for all the PI controllers.
137/188
Figure 10-1: Nyquist plots of the first (left) and second (right) feedback loop for the transfer
function model linearized at the first operating point DP1.
Figure 10-2: Nyquist plots of the first (left) and second (right) feedback loop for the transfer
function model linearized at the second operating point DP4.
In order to take into account the effects of the loop interactions, MIMO Nyquist
stability criteria are also tested, as suggested by Skogestad and Postlethwaite (2005).
The determinant of the identity matrix plus the open loop transfer function with the
controllers, det(I+L(s)) should not make any encirclements of the origin if L(s) is
stable. In this case we have a 2 × 2 system Gp, with a diagonal controller Gc. The open
loop transfer function is defined as follows:
1 + G p11Gc1 G p12 Gc 2
det(1 + L( s )) = det = (1 + G p11Gc1 )(1 + G p 22 Gc 2 ) − G p12 Gc 2 G p 21Gc1
G p 21Gc1 1 + G p 22 Gc 2
(10-10)
The determinant was calculated for the transfer functions with first order pade
approximations for the time delays, as defined in Equation (8-13). The Nyquist plots
for the transfer functions determined at the first and second operating points are
presented in Figure 10-3 and in Figure 10-4, respectively. Since neither of them
encircle the origin, the process with controllers is stable.
Figure 10-3: Nyquist plots of det (I+L(s)) for the transfer function model linearized at the first
operating point DP1.
139/188
Figure 10-4: Nyquist plots of det (I+L(s)) for the transfer function model linearized at the second
operating point DP4.
Figure 10-5: Responses of the rich electrolyte copper concentration to ±5% changes in F(PLSS),
F(PLSP), c(PLS), c(LE) and setpoint of the c(LO) with open control loops (OL) and under a
closed FB1 control loop (PI).
The interactions at operating point DP4 for the rich electrolyte copper concentration
were stronger than the interactions at operating point DP1. The interaction between
loop FB1 and the rich electrolyte copper concentration was significant. When the FB1
loop was closed, the responses to 5% changes in the PLS series and parallel flow
rates, F(PLSS) and F(PLSP), and PLS copper concentration, c(PLS), were almost
twice as large as those with both control loops open. The FB1 loop had almost no
effect on the response to the change in the lean electrolyte copper concentration
c(LE), as can be seen from Figure 10-6. When FB1 loop was closed, a 5% change in
the c(LO) setpoint caused a significant and very asymmetric change in the rich
electrolyte copper concentration, c(RE), via manipulations in the loaded organic flow
rate, F(LO). Therefore, a decoupler could be included in the control strategy.
141/188
Figure 10-6: Responses of the rich electrolyte copper concentration to ±5% changes in F(PLSS),
F(PLSP), c(PLS), c(LE) and setpoint of the c(LO) with open control loops (OL) and under a
closed FB1 control loop (PI).
Figure 10-7: Responses of the load organic copper concentration to ±5% changes in F(PLSS),
F(PLSP), c(PLS), c(LE) and setpoint of the c(RE) with open control loops (OL) and under a
closed FB2 control loop (PI).
The interaction between loop FB2 and the loaded organic copper concentration was
relatively small at operating point DP4. When the FB1 loop was closed, the responses
to the 5% changes in the PLS series and parallel flow rates, F(PLSS) and F(PLSP),
and PLS copper concentration, c(PLS), were similar to the responses with open
control loops. The FB1 loop decreased the effect of the change in the lean electrolyte
copper concentration c(LE), but turned it into an inverse response, as can be seen
from Figure 10-8. When the FB2 loop was closed, a 5% change in the c(RE) setpoint
caused as large an effect as the changes in the flow rate to the loaded organic copper
concentration, c(LO). Because the interactions of the FB2 loop are relatively small, a
decoupler is not necessary.
Figure 10-8: Responses of the load organic copper concentration to ±5% changes in F(PLSS),
F(PLSP), c(PLS), c(LE) and setpoint of the c(RE) with open control loops (OL) and under a
closed FB2 control loop (PI).
143/188
Table 10-1: The integral of absolute error (IAE) for loaded organic copper concentration with
disturbances at the first operating point.
Figure 10-9: Loaded organic copper concentration with a +5% change in F(PLSS) and F(PLSP),
under open loop control (OL), PI controllers (PI), and PI controllers with feedforward
compensators (PIFF) at the first operating point.
Similar results are obtained for the PLS copper concentration, c(PLS) at sampling
period [2000, 3000]. The lean electrolyte copper concentration disturbances at
sampling period [3000, 4000] are satisfactorily rejected with both PI controllers and
additional feedforward compensators, as can be seen from Figure 10-10:
145/188
Figure 10-10: Loaded organic copper concentration with a +5% change in c(PLS) and c(LE),
under open loop control (OL), PI controllers (PI), and PI controllers with feedforward
compensators (PIFF) at the first operating point.
Figure 10-11: Loaded organic copper concentration with a +5% change in c(PLS)&F(PLSS),
c(PLS)&F(PLSP), and c(PLS)&c(LE), under open loop control (OL), PI controllers (PI), and PI
controllers with feedforward compensators (PIFF) at the first operating point.
The disturbance rejection results evaluated at the second operating point DP4 are
presented in Table 10-2, Figure 10-12, Figure 10-13 and Figure 10-14.
Disturbance rejection with the first PI controller (FB1) was very good. Opening the
FB1 loop and closing the FB2 loop (FB2) decreases the performance to a similar level
as for the open loop case (OL). The disturbance rejection performance was increased
with both loops closed (PI), and further improved by adding the feedforward
compensators (PIFF), as can be seen from Table 10-2.
146/188
Table 10-2: The integral of absolute error (IAE) for loaded organic copper concentration with
disturbances at the second operating point.
In the following visual analysis the open loop responses are compared to the PI-
controller performance with both loops closed (PI), and to the performance of the two
PI-control loops with four feedforward compensators.
The ±5% disturbances to the PLS series and parallel flow rates were effectively
rejected with the PI controllers, as can be seen from Figure 10-12. Even better
performance was gained by adding feedforward compensators to the control system.
For example, with a +5% change in F(PLSS) the integral of the absolute error for
open loop control is 45.2, whereas PI control decreases the index to 4.4 and PI control
with feedforward compensator down to 2.2. For a PLS parallel flow rate disturbance
the rejection is even better: with PI controllers the IAE index decreases from 49.2 to
5.0. Adding the feed forward compensators improves the index to 1.4.
Figure 10-12: Loaded organic copper concentration with a +5% change in F(PLSS) and
F(PLSP), under open loop control (OL), PI controllers (PI), and PI controllers with feedforward
compensators (PIFF) at the second operating point.
The disturbances in the copper concentrations of PLS and lean electrolyte are rejected
satisfactorily, as can be seen from Figure 10-13. Adding the feedforward
compensators decreases the IAE index for the PLS change from 10.8 to 4.5, and for
the lean electrolyte copper concentration from 1.9 to 0.6.
147/188
Figure 10-13: Loaded organic copper concentration with a +5% change in c(PLS) and c(LE),
under open loop control (OL), PI controllers (PI), and PI controllers with feedforward
compensators (PIFF) at the second operating point.
Figure 10-14: Loaded organic copper concentration with a +5% change in c(PLS)&F(PLSS),
c(PLS)&F(PLSP), and c(PLS)&c(LE), under open loop control (OL), PI controllers (PI), and PI
controllers with feedforward compensators (PIFF) at the second operating point.
feedforward compensators (PIFF) improved the result considerably for the lean
electrolyte copper concentration disturbance, as can be seen from Table 10-3.
Table 10-3: The integral of absolute error (IAE) for the rich electrolyte copper concentration
with disturbances at the first operating point.
In the following visual analysis the open loop responses are compared to the PI-
controller performance with both loops closed (PI), and to the performance of the two
PI-control loops with four feedforward compensators.
The feedforward compensators fasten the disturbance rejection, although they cause
higher peaks than the PI controllers alone, as shown for the PLS flow rate
disturbances, F(PLSS) and F(PLSP), in Figure 10-15. The deviation from the setpoint
is smaller with feedforward compensators, as shown in Table 10-3.
Figure 10-15: Rich electrolyte copper concentration with a +5% change in F(PLSS) and
F(PLSP), under open loop control (OL), PI controllers (PI), and PI controllers with feedforward
compensators (PIFF) at the first operating point.
For the PLS copper concentration, c(PLS) at sampling period [2000, 3000], the
disturbance rejection performance is similar to the previous ones. The lean electrolyte
copper concentration disturbance rejection at sampling period [3000, 4000], however,
is significantly improved with the feedforward compensation, as shown in Figure
10-16.
149/188
Figure 10-16: Rich electrolyte copper concentration with a +5% change in c(PLS) and c(LE),
under open loop control (OL), PI controllers (PI), and PI controllers with feedforward
compensators (PIFF) at the first operating point.
Figure 10-17: Rich electrolyte copper concentration with a +5% change in c(PLS)&F(PLSS),
c(PLS)&F(PLSP), and c(PLS)&c(LE), under open loop control (OL), PI controllers (PI), and PI
controllers with feedforward compensators (PIFF) at the first operating point.
The disturbance rejection results evaluated in the second operating point DP4 are
presented in Table 10-4, Figure 10-18, Figure 10-19 and Figure 10-20.
The disturbance rejection with the second PI controller (FB2) was excellent. Opening
the FB2 loop and closing the FB1 loop (FB1) has a drastic effect on the disturbance
rejection due to high interaction between the loops and, for some cases, the
performance drops to worse levels than for the open loop case (OL). The setpoint
tracking performance with both control loops closed (PI) was worse than the FB2
loop alone due to the interactions. Thus, adding the feedforward compensators (PIFF)
improved the result considerably for the lean electrolyte copper concentration
disturbance, as can be seen from Table 10-4.
150/188
Table 10-4: The integral of absolute error (IAE) for rich electrolyte copper concentration with
disturbances at the second operating point.
In the following visual analysis the open loop responses are compared to the PI-
controller performance with both loops closed (PI), and to the performance of the two
PI-control loops with four feedforward compensators.
The feedforward compensators fasten the disturbance rejection, although causing
higher peaks than the PI controllers alone, as shown for the PLS flow rate
disturbances F(PLSS) at sampling period [0, 1000] and for F(PLSP) at sampling
period [1000, 2000] in Figure 10-18.
Figure 10-18: Rich electrolyte copper concentration with a +5% change in F(PLSS) and
F(PLSP), under open loop control (OL), PI controllers (PI), and PI controllers with feedforward
compensators (PIFF) at the second operating point.
The disturbance rejection performance for the PLS copper concentration c(PLS) at
sampling period [2000, 3000] is similar to the previous ones. The lean electrolyte
copper concentration disturbance rejection at sampling period [3000, 4000], however,
is significantly improved with the feedforward compensation, as shown in Figure
10-19.
151/188
Figure 10-19: Rich electrolyte copper concentration with a +5% change in c(PLS) and c(LE),
under open loop control (OL), PI controllers (PI), and PI controllers with feedforward
compensators (PIFF) at the second operating point.
The simultaneous disturbances are rejected the best by adding the feedforward
compensators with the PI controllers, as can be seen from Figure 10-20, and Table
10-4.
Figure 10-20: Rich electrolyte copper concentration with a +5% change in c(PLS)&F(PLSS),
c(PLS)&F(PLSP), and c(PLS)&c(LE), under open loop control (OL), PI controllers (PI), and PI
controllers with feedforward compensators (PIFF) at the second operating point.
152/188
Table 10-5: The integral of absolute error (IAE) for the loaded organic copper concentration
with setpoint tracking at the first operating point.
Figure 10-21: Loaded organic copper concentration with +5% changes in c(LO) setpoint, c(RE)
setpoint, and simultaneously to both setpoints, under open loop control (OL), PI controllers (PI),
and PI controllers with feedforward compensators (PIFF) at the first operating point.
The setpoint tracking performances at the second operating point DP4 are numerically
presented in Table 10-6 for all the controller combinations, and illustrated in Figure
10-22 for the open loop (OL), two PI controller (PI) and two PI controllers with four
feedforward compensators (PIFF).
The setpoint tracking with the first PI controller (FB1) was excellent. Opening the
FB1 loop and adding the second PI controller (FB2) decrease the performance to a
similar level as for the open loop case (OL). The setpoint tracking performance was
minimally decreased by closing both loops (PI). Adding the feedforward
compensators (PIFF) did not affect the performance, as can be seen from Table 10-6.
154/188
Table 10-6: The integral of absolute error (IAE) for the loaded organic copper concentration
with set point tracking at the second operating point .
The setpoint tracking for the loaded organic copper concentration is successful with
both PI controllers. If the controller was more aggressive, this would result in
disturbances in the rich electrolyte copper concentration. Changes in the rich
electrolyte copper concentration setpoint cause only small disturbances, although in
normal operation both setpoints are either raised or decreased simultaneously. The
setpoint tracking during the simultaneous setpoint changes results in slightly more
aggressive responses, as can be seen from Figure 10-22.
Figure 10-22: Loaded organic copper concentration with +5% changes in c(LO) setpoint, c(RE)
setpoint, and simultaneously to both setpoints, under open loop control (OL), PI controllers (PI),
and PI controllers with feedforward compensators (PIFF) at the second operating point.
The setpoint tracking performance with both control loops closed (PI) improved the
result compared to the FB1 case, as can be seen from Table 10-7.
Table 10-7: The integral of absolute error (IAE) for the rich electrolyte copper concentration
with set point tracking at the first operating point.
The setpoint change for the loaded organic copper concentration c(LO) at sampling
period [7000, 8000] causes small peaks of less than 20% in the rich electrolyte copper
concentration due to the control loop interactions. The tracking of the setpoint
changes in the rich electrolyte copper concentration c(RE) is is adequately fast and
effective in the single change at sampling period [8000, 9000] and the simultaneous
change at sampling period [9000, 10000] cases, as can be observed from Figure
10-23.
Figure 10-23: Rich electrolyte copper concentration with +5% changes in c(LO) setpoint, c(RE)
setpoint, and simultaneously to both setpoints, under open loop control (OL), PI controllers (PI),
and PI controllers with feedforward compensators (PIFF) at the first operating point.
The setpoint tracking performances at the second operating point DP4 with different
controllers are presented in Table 10-8 and in Figure 10-24.
The setpoint tracking with the second PI controller (FB2) was excellent. Opening the
FB2 loop and closing the FB1 loop (FB1) have a drastic effect on disturbance
rejection due to the high interaction between the loops and, for some cases, the
performance drops to worse levels than for the open loop case (OL). The setpoint
tracking performance with both control loops closed (PI) improved the result
compared to the FB1 loop closed, as shown in Table 10-8.
156/188
Table 10-8: The integral of absolute error (IAE) for the rich electrolyte copper concentration
with disturbances and set point tracking at the second operating point.
The setpoint change for the loaded organic copper concentration c(LO) at sampling
period [7000, 8000] causes peaks of about 35% in the rich electrolyte copper
concentration due to the high loop interactions. The tracking of the setpoint changes
in the rich electrolyte copper concentration c(RE) is adequately fast and effective in
the single change at sampling period [8000, 9000]. The simultaneous setpoint change
in the loaded organic copper concentration at sampling period [9000, 10000] causes
slightly more oscillating behaviour, despite the adequately good setpoint tracking, as
can be seen from Figure 10-24.
Figure 10-24: Rich electrolyte copper concentration with +5% changes in c(LO) setpoint, c(RE)
setpoint, and simultaneously to both setpoints, under open loop control (OL), PI controllers (PI),
and PI controllers with feedforward compensators (PIFF) at the second operating point.
disturbance variables, one at a time and simultaneously. The changes in the copper
concentrations and flow rates were well rejected with the PI controllers. Adding the
feedforward compensators improved the disturbance rejection significantly, by around
50 – 70% for the loaded organic copper concentration with the disturbances in the
PLS copper concentration and flow rates, and by around 80% for the rich electrolyte
copper concentration with the disturbance in the lean electrolyte copper concentration.
158/188
)
where yr is the reference trajectory, y is the output prediction and u is the input value,
k is the present moment and d is the discrete dead time. The tuning parameters are the
prediction horizon (n2-n1), the control horizon nu, and the weights γy and γu. The
output weight γy is punishing the error between the output and the reference trajectory,
and γu is punishing the changes in the manipulated variable.
The current control action is obtained by solving, online at each sampling time, a
finite horizon open-loop optimal control problem using the current state as the initial
state. The optimization yields an optimal control sequence, and the first control in this
sequence is applied to the plant.
All the MPC methods share the following four basic elements. The process model is
used to predict the future outputs of the process. The future outputs are compared to
the reference trajectory, and the future errors and the costs of the control actions to be
made are optimized. Both the cost function and constraints are considered in the
optimization. The future inputs are calculated, and the next control action is
performed. The strategy and philosophy of the implementation of the elements differs
between the different methods, but the general structure follows the scheme presented
in Figure 11-1. (Camacho and Bordons 1999, Henson, 1998, Mayne, 2000, Rawlings,
2000, Maciejovski, 2002).
159/188
Figure 11-1: The general strategy of MPC. (Camacho and Bordons 1999).
The first MPC techniques were developed in the 1970s because conventional single-
loop controllers were unable to satisfy the increasingly stringent performance
requirements. MPC is well suited for high performance control of constrained
multivariable processes. The current generation of industrial model predictive
controllers is based on the assumption of process linearity, because this simplifies
model development and controller design. The empirical dynamic models are
identified from test data, and the stability of the tuning is assured by testing the
scheme with closed-loop simulations. (Morari and Lee, 1999, Qin and Badgwell,
2003)
MPC technology has been used extensively in the refining and petrochemical
industries. During the past decade MPC strategies have also been applied to other
areas. Adaptation was one of the motivations for MPC, and there is a strong market
incentive for a self-tuning model predictive controller. (Mayne, 2000, Qin and
Badgwell, 2003).
prediction horizon (n2-n1), control horizon (nu), and cost function weights for the
controlled variable weights (γy) and manipulated variable rate weights (γu). Hard
constraints were not assigned. The initial choices for the parameters were the
following: the prediction horizon was the longest settling time of the manipulated
variable-controlled variable pairs, and the control horizon was longer than the longest
dead time of the manipulated variable-controlled variable pairs. The rich electrolyte
copper concentration had a larger weight than the loaded organic copper
concentration in order to emphasize the importance of the end product quality. The
manipulated variable rate weights were tuned to avoid large changes and oscillating
behaviour. For the fine tuning of the horizons and weights, minimization of the iae
index was used. In the following sections the disturbance rejection and setpoint
tracking performance of the model predictive controller is compared to that of the PI
controllers.
Table 11-1: The integral of absolute error (IAE) for the loaded organic copper concentration
with disturbances and set point tracking at the first operating point.
Figure 11-2: Loaded organic copper concentration with a +5% change in F(PLSS) and F(PLSP),
under open loop control (OL), PI controllers (PI), and model predictive controller (MPC) at the
first operating point.
As with the PLS flow rate disturbances, the PLS copper concentration disturbance at
sampling period [2000, 3000] is more effectively rejected with MPC than with PI
controllers. However, a disturbance in the lean electrolyte copper concentration
causes larger changes in the loaded organic copper concentration with MPC than with
PI controllers, as shown in Figure 11-3. This is due to the MPC tuning, which favours
162/188
disturbance rejection in the rich electrolyte copper concentration over the loaded
organic copper concentration.
Figure 11-3: Loaded organic copper concentration with a +5% change in c(PLS) and c(LE),
under open loop control (OL), PI controllers (PI), and model predictive controller (MPC) at the
first operating point.
All the combined disturbances are rejected more effectively with MPC than with PI
controllers, as can be seen from Figure 11-4. For the last combination of PLS and lean
electrolyte copper concentration disturbances at sampling period [6000, 7000], MPC
has a lower integral of the absolute error value than PI controllers (IAE 20.3 and 21.0
correspondingly), as shown in Table 11-1.
Figure 11-4: Loaded organic copper concentration with a +5% change in c(PLS)&F(PLSS),
c(PLS)&F(PLSP), and c(PLS)&c(LE), under open loop control (OL), PI controllers (PI), and
model predictive controller (MPC) at the first operating point.
Table 11-2: The integral of absolute error (IAE) for the rich electrolyte copper concentration
with disturbances at the first operating point.
Figure 11-5: Rich electrolyte copper concentration with a +5% change in F(PLSS) and F(PLSP),
under open loop control (OL), PI controllers (PI), and model predictive controller (MPC) at the
first operating point.
The disturbances in the PLS copper concentration c(PLS) at sampling period [2000,
3000], and the lean electrolyte copper concentration, c(LE) at sampling period [3000,
4000], are especially well rejected with MPC, as shown in Figure 11-6. The
performance for the PLS copper concentration with MPC is four times better than
with PI controllers (IAE 3.3 and 16.0 correspondingly), and the for lean electrolyte
copper concentration almost twice as good (IAE 28.5 and 40.8 correspondingly).
164/188
Figure 11-6: Rich electrolyte copper concentration with a +5% change in c(PLS) and c(LE),
under open loop control (OL), PI controllers (PI), and model predictive controller (MPC) at the
first operating point.
The simultaneous disturbances are rejected far better with MPC than with PI
controllers, as can be seen from Figure 11-7, and Table 11-2.
Figure 11-7: Rich electrolyte copper concentration with a +5% change in c(PLS)&F(PLSS),
c(PLS)&F(PLSP), and c(PLS)&c(LE), under open loop control (OL), PI controllers (PI), and
model predictive controller (MPC) at the first operating point.
• simultaneous with the same signs for the loaded organic copper concentration
and rich electrolyte copper concentration, c(LO)&c(RE)
The measure used to compare the controller combinations was the integral of the
absolute error (IAE) between the setpoints of the controlled variables and the outputs
of the controlled variables under control. The results for the loaded organic copper
concentration are presented in Section 11.4.1 and for the rich electrolyte copper
concentration are presented in Section 11.4.2.
11.4.1 Loaded organic copper concentration
The loaded organic copper concentration setpoint tracking performance at operating
point DP1 with different control schemes is elaborated in the following section, with
numerical and visual examples in Table 11-3 and Figure 11-8.
The results obtained with the model predictive controller were better for setpoint
tracking than with the PI controllers, as can be seen from Table 11-3. The controller
interaction is significantly decreased with MPC; for example, the setpoint tracking of
the rich electrolyte copper concentration c(RE) has a value of 4.8 with MPC and of
15.7 with PI controllers. Within the MPC structure the rich electrolyte copper
concentration setpoint tracking had a higher coefficient than the loaded organic
copper concentration, and thus here the loop interaction counteracts the good result
for the loaded organic copper concentration.
Table 11-3: The integral of absolute error (IAE) for the loaded organic copper concentration
with set point tracking at the first operating point.
Adaptation to the new loaded organic setpoint [7000 8000] is fast with only small
overshoot. The simultaneous setpoint change for both of the controlled variables
results in similar responses between the MPC and PI controller. However, as can be
seen from the previous table, MPC is slightly better than the PI controllers (29.1 and
30.2 correspondingly). The effect of the rich electrolyte copper concentration setpoint
change is also far smaller with the model predictive controller than with the PI
controllers (IAE 4.8 and 15.7, respectively).
166/188
Figure 11-8: Loaded organic copper concentration with a +5% changes in c(LO) setpoint, c(RE)
setpoint, and simultaneously to both setpoints, under open loop control (OL), PI controllers (PI),
and model predictive controller (MPC) at the first operating point.
Table 11-4: The integral of absolute error (IAE) for the rich electrolyte copper concentration
with set point tracking at the first operating point.
The setpoint tracking is more effective with MPC than with PI controllers, as
illustrated in Figure 11-9. The loaded organic copper concentration setpoint tracking
[7000 8000] causes a smaller disturbance to the rich electrolyte copper concentration
with MPC than with PI control (IAE 7.7 and 19.9, respectively). The change in the
setpoint of the rich electrolyte copper concentration at sampling period [8000, 9000]
is varied faster with MPC than with PI controllers (IAE 58.5 and 64.4, respectively).
167/188
MPC performs especially well with the simultaneous setpoint change in both
controlled variables: the IAE index is 58.4, whereas for PI controllers the index is
81.5.
Figure 11-9: Rich electrolyte copper concentration with +5% changes in c(LO) setpoint, c(RE)
setpoint, and simultaneously to both setpoints, under open loop control (OL), PI controllers (PI),
and model predictive controller (MPC) at the first operating point.
∑c sp (i ) − c(i )
AAE = i =1
(12-2)
N
The industrial measurements represent the manual control strategy. The setpoints for
the controlled variables are determined as the one-day moving average of the
corresponding industrial online measurement.
The inputs to the simulator with the controllers for the single-input single-output and
multi-input-multi-output control strategies are the industrial measurements and varied
parameters for the equilibrium isotherm, efficiencies and recycle correction, as
described in Chapter 7. The PI controllers and feedforward (FF) compensators
described in Chapter 10, and the model predictive controller (MPC) with the
feedforward compensator for the lean electrolyte copper concentration described in
Chapter11, are utilized. The setpoints for the controlled variables are determined by
solving the optimization problem with the offline process data, as described in Section
9.7.
In the optimization, the minimum and maximum flow rates of the industrial data are
considered as additional constraints. Also the organic to aqueous ratios are determined
169/188
on the basis of the maximum value for the extraction and the minimum value for the
stripping from the industrial data.
No assumptions are made about the operation of the electrowinning, but the lean
electrolyte copper concentration measurements are used as such. Such an assumption
might affect the control results negatively, since the rich electrolyte control loop has
to do extra work to stabilize the disturbances in the lean electrolyte copper
concentration. If an electrowinning model existed, the electrolyte copper
concentrations could be stabilized at the desired levels.
The control results are presented for the production increase in Section 12.2 and for
the variation decrease in Section 12.3.
Table 12-1: The production increase with the different control strategies compared to the manual
control strategy.
Production Production
First test data set second test data set
PI +4.93% +2.77%
PI+FF +4.92% +2.81%
MPC +4.93% +2.78%
MPC+FF +4.93%
For the first data set, the production under the PI controllers with the feedforward
compensator is presented in Figure 12-1, and under MPC control with the c(LE)
feedforward compensator in Figure 12-2. The production is clearly higher and has less
variation compared to that for the manual operating practice. The visual comparison
shows that the model predictive controller also causes less variation in the production
compared to the PI controllers with the feedforward compensators.
170/188
Figure 12-1: Copper production for the first test data set: industrial measurement (dotted) and
PI+FF controlled in the simulation environment (solid).
Figure 12-2: Copper production for the first test data set: industrial measurement (dotted) and
MPC+FF controlled in the simulation environment (solid).
For the second data set, the production is more stable with the MPC controller, as
shown in Figure 12-4, compared to the PI controlled production presented in Figure
12-3. However, the production under MPC has more small variation, whereas the
production with PI controllers has slower dynamics.
171/188
Figure 12-3: Copper production for the second test data set: industrial measurement (dotted) and
PI controlled in the simulation environment (solid).
Figure 12-4: Copper production for the first test data set: industrial measurement (dotted) and
MPC controlled in the simulation environment (solid).
decreasing the average absolute error by around 30-40% for the loaded organic copper
concentration. Adding the feedforward compensators decreased the average absolute
error for the PI control strategy to 38% for the loaded organic copper concentration.
Table 12-2: The variation decrease for the loaded organic copper concentration with the different
control strategies compared to the manual control strategy.
For the first test data set, the PI controller with the feedforward compensators better
tracks the setpoint for the loaded organic copper concentration, as shown in Figure
12-5, than the MPC with the feedforward compensator, as presented in Figure 12-6.
The optimization for this data set reduces the setpoint of the loaded organic copper
concentration to a lower level than the measurement, but the flow rate is higher and
therefore more copper is transferred from the pregnant leach solution to the organic
solution.
Figure 12-5: Loaded organic copper concentration for the first test data set: industrial
measurement (dotted, black), PI+FF controlled in the simulation environment (solid, blue), and
setpoint (red, dashed).
173/188
Figure 12-6: Loaded organic copper concentration for the first test data set: industrial
measurement (dotted, black), MPC+FF controlled in the simulation environment (solid, blue),
and setpoint (red, dashed).
For the second test data set, the setpoint tracking performance of both controllers are
good for the loaded organic copper concentration, as illustrated for PI controllers in
Figure 12-7 and for the MPC in Figure 12-8. The setpoint is slightly lower than the
industrial measurement and the organic flow rate therefore higher in order to achieve
higher copper extraction.
Figure 12-7: Loaded organic copper concentration for the second test data set: industrial
measurement (dotted, black), PI controlled in the simulation environment (solid, blue), and
setpoint (red, dashed).
174/188
Figure 12-8: Loaded organic copper concentration for the second test data set: industrial
measurement (dotted, black), MPC controlled in the simulation environment (solid, blue), and
setpoint (red, dashed).
Table 12-3: The variation decrease for the rich electrolyte copper concentration with the
different control strategies compared to the manual control strategy.
For the first test data set, the setpoint tracking for the rich electrolyte copper
concentration is more effective with the MPC with feedforward compensator, as
shown in Figure 12-10, than with the PI controllers with feedforward compensators,
as presented in Figure 12-9. The rich electrolyte copper concentration setpoint is
higher than the measurement, but the flow rate is slightly lower.
175/188
Figure 12-9: Rich electrolyte copper concentration for the first test data set: industrial
measurement (dotted, black), PI+FF controlled in the simulation environment (solid, blue), and
setpoint (red, dashed).
Figure 12-10: Rich electrolyte copper concentration for the first test data set: industrial
measurement (dotted, black), MPC+FF controlled in the simulation environment (solid, blue),
and setpoint (red, dashed).
For the second test data set, the rich electrolyte copper concentration setpoint tracking
is successful with both controllers, as shown for PI controllers in Figure 12-11 and for
MPC in Figure 12-12. The setpoint is almost at the same level as the measurement.
For this data set, the maximum flow rate constraints restrict the increase in
production.
176/188
Figure 12-11: Rich electrolyte copper concentration for the second test data set: industrial
measurement (dotted, black), PI controlled in the simulation environment (solid, blue), and
setpoint (red).
Figure 12-12: Rich electrolyte copper concentration for the second test data set: industrial
measurement (dotted, black), MPC controlled in the simulation environment (solid, blue), and
setpoint (red).
The results could be improved by adding an electrowinning model that would provide
more realistic lean electrolyte copper concentrations for the solvent extraction model.
By stabilizing the rich electrolyte copper concentration, the lean electrolyte copper
concentration would also have less variation and the production could be increased.
178/188
14 CONCLUSIONS
Novel dynamic models and a novel simulation tool for the industrial copper solvent
extraction process have been developed in this work. The models facilitate studies on
the dynamic behaviour and development of control system for industrial copper
solvent extraction plants. The model structure is divided into unit processes, and
therefore can be easily modified to represent different plant configurations. With the
dynamic models the cause and effect relationships of the process can be easily
illustrated, and an understanding of the process principles taught. A simulator, based
on the dynamic models, could be used as an operator’s training tool and as a plant
optimization tool.
The dynamic models are necessary for the development of a control system. In this
work, two novel control strategies have been designed. The single input-single output
and multi input – multi output control strategies on the stabilizing control level, and
the optimization algorithm on the optimization level, were based on linearized process
models. Due to the simple controller structures based on phenomenological process
models, the control system should be easily accepted by industrial partners. Testing
the control system in the simulation environment showed a significant increase in the
performance and profitability of the operation. This greatly encourages the testing of
the control system in an industrial plant.
In the future, the solvent extraction models could be extended by modeling the mixer
as series of two or more mixing units and adding back-mixing models for the settlers.
In most of the copper solvent extraction plants the mixer is series of two to three
mixing units and this addition might increase the accuracy of the modeling. The
settler time delays could be set to be time variant, and the model parameters could be
estimated from the online measurements directly.
Since the operating data from industrial plants has often problems with accuracy/ non-
existents of some measurements, it would be beneficial to use a pilot plant to collect
data, especially online data of variables that in this study were available only offline.
Thus, for further model development it is suggested to collect data from a well
instrumented pilot plant under normal operating conditions in different operating
points. Including online measurement copper concentrations, pH, acidity and reagent
volume percent from all the unit processes of the plant would enhance the modeling
as well as the control.
The dynamic solvent extraction model should be combined with a dynamic
electrowinning model. This would enable even more realistic studies of the process
chain behaviour, and enable real time optimization of a network of several solvent
extraction and electrowinning processes. Since electrowinning is a more complicated
process than solvent extraction, a simple approach, such as adaptive data based
models, could provide the easiest basis for further modeling and control studies.
One critical issue in industrial copper solvent extraction plants is the control of the
impurity level, for example of iron, manganese and chloride levels. Since an x-ray
analyzer is also able to measure concentration of these species, the impurities could be
added to the control system. Dynamic models that include the different mass transfer
paths of each species should therefore be developed.
A ratio controller for the parallel and series PLS flow should be added to the control
structure in order to optimize copper extraction from the total available PLS. If an
180/188
electrowinning model is added, a new control structure should be designed. For the
combined model the electric current in electrowinning should be an additional
manipulated variable. Model predictive control could be a convenient control
algorithm for the combined system due to the good interaction and time delay
handling properties. Application of nonlinear model predictive control could further
enhance the control performance due to the ability to explicitly take into account the
process nonlinearities. However, the performance of the nonlinear MPC algorithm
should be compared to the performance of the linear MPC algorithm to justify the
development and maintenance costs associated with industrial use of a more complex
process model.
With the dynamic models, the simulation tool and the control system, a plant-wide
operator support system can be developed and the performance and profitability of an
industrial copper solvent extraction process significantly improved.
181/188
15 REFERENCES
Abdeltawab, A.A., Nii, S., Kawaizumi, F., and Takahashi, K. (2002), Separation of
La and Ce with PC-88A by counter-current mixer-settler extraction column,
Separation and Purification Technology, 26, pp 265-272.
Aminian, H., Bazin, C., Hodouin, D., Jacob, C., (2000), Simulation of a SX-EW pilot
plant, Hydrometallurgy, 56, pp. 13-31.
Aminian, H., Bazin, C., Hodouin, D. (1998), Residence time distriburion in SX-EW
equipment, Int. J. Miner. Process., 54, pp. 235-242.
Anon (2000), The Acorga Technical Library: The McCabe-Thiele Diagram And
Theoretical Design Aspects For Solvent Extraction Plants,
http://www.acorga.com/htdocs/pdf/mc.pdf, ref. 5.10.2006.
Bartos, P.J.(2002), SX-EW copper and the technology cycle, Resources Policy, 28,
pp. 85-94.
Bazin, C., Hodouin, D., Zouadi, M. (2005), Data reconciliation and equilibrium
constant estimation: Application to copper solvent extraction, Hydrometallurgy, 80,
pp. 43-53.
Bergh, L.G., Lucic, E.M., and Zuleta, F. (2006), Supervisory control project of a
copper solvent extraction pilot plant, In: Proceedings of the IFAC workshop
MMM’2006 – Automation in Mining, Mineral and Metal Industry, (eds. Cierpisz, S.,
Miskiewicz, K., Heyduk, A.), Cracow, IFAC, pp.127-131.
Bergh, L.G., and Yianatos, J.B. (2001), Current Status and Limitations of Copper
SX/EW Plants Control, Minerals Engineering, 14, pp. 975 – 985.
182/188
Bergh, L.G., Jämsä-Jounela, S.-L., Hodouin, D., (2001) , State of the art in copper
hydrometallurgic processes control, Control Engineering Practice, 9, pp. 1007-1012.
Biswas, A.K., and Davenport, W.G. (1994), Extractive metallurgy of copper, 3rd ed.,
Pergamon, Oxford, pp. 500.
Galvez, E.D., Vega, C.A., Swaney, R.E., Cisternas, L.A. (2004), Design of solvent
extraction circuit schemes, Hydrometallurgy, 74, pp. 19–38.
Glad, T., and Ljung, L. (2000), Control Theory, Multivariable and Nonlinear
Methods, Taylor & Francis, London, pp. 27, 219, 308 (467).
Habashi, F. (1999), The future of extractive metallurgy, CIM Bulletin, 92, pp.161-
165.
Henson, M. A. (1998), Nonlinear model predictive control: current status and future
derections, Computers and Chemical Engineering, 23, pp. 187 – 202.
Hillier, F.S. and Lieberman, G.J. (2001), Introduction to operations research, 7th ed.
McGraw-Hill, Boston, pp. 24-42 (1214).
183/188
Hodouin, D., Jämsä-Jounela, S. -L., Carvalho, M. T. and Bergh, L.(2001), State of the
art and challenges in mineral processing control, Control Engineering Practice, 9, pp.
995-1005
Hodouin, D., Bazin, C., Gagnon, E., and Flament, F. (2000), Feedforward-feedback
predictive control of a simulated flotation bank, Powder Technology, 108, pp. 173 –
179.
Hoh Y.C., Ju, S.J. and Chiu T.M. (1989), Effect of internal recycle on mixer-settler
performance, Hydrometallurgy, 23, pp. 105-118.
Ingham, J., Dunn, I.J., Heinzle, E., Prenosil, J.E., (1994), Chemical Engineering
Dynamics - Modelling with PC Simulation, VCH Verlagsgesellschaft mbH, Weinham,
pp. 166 – 193 (701).
Jenkins, J., Davenport, W.G., Kennedy, B., Robinson, T., (1999), Electrolytic copper
– leach, solvent extraction and electrowinning world operating data, In: Proceedings
of the Copper 99 – Cobre 99 International Conference – Volume IV, (eds. Young,
S.K., Dreisinger, D.B. Hackl, R.P. , Dixon D.G.), Phoenix, TMS, Warrendale, pp.
493-566.
Jämsä-Jounela, S.-L., (2001), Current status and future trends in the automation of
mineral and metal processing, Control Engineering Practice 9, pp. 1021-1035.
Komulainen, T., Doyle, F.J., Rantala, A., Jämsä-Jounela, S.-L (2007), Control of an
industrial copper solvent extraction process, Journal of Process Control, (submitted).
Komulainen, T., Pekkala, P., Rantala, A., Jämsä-Jounela, S.-L (2006), Dynamic
modelling of an industrial copper solvent extraction process. , Hydrometallurgy, 81,
pp. 52-61 .
184/188
Kordosky, G., Virnig, M., Boley, B. (2006), Equilibrium copper strip points as a
function of temperature and other operating parameters: Implications for commercial
copper solvent extraction plants, Tsinghua Science and Technology, 11, pp. 160-164.
Kordosky, G.A., Olafson, S., and Hahn, C. (2000), Iron Control Strategies in Copper
Solvent Extraction Plants, The Case for Wash Stage, In: Proceedings of ALTA
Copper-6 Conference, Perth Australia, ALTA Metallurgical Services, (CD).
Kordosky, G.A., Olafson, S.M., Lewis, R.G., and Deffner, V.L. (1987), A State of the
Art Discussion on the Solvent Extraction reagents Used for the recovery of Copper
from Dilute Sulfuric Acid Leach Solutions, Separation science and technology, 22,
pp. 215 – 232.
Kreyzig, E. (1999), Advanced Engineering Mathematics, 8th ed., John Wiley & Sons
Inc., New York, pp.994-1001 (1156).
Ljung, L. 2006, System Identification Toolbox for use with Matlab – User’s Guide, 6th
ed., http://www.mathworks.com/access/helpdesk/help/pdf_doc/ident/ident.pdf, The
MathWorks Inc., p. 228.
Maciejowski, J.M., (2002), Predictive control with constraints, Prentice Hall, Harlow,
England, pp. 352.
MathWorks, (2006), Signal processing toolbox for use with Matlab – User’s Guide,
6th ed., http://www.mathworks.com/access/helpdesk/help/pdf_doc/signal/signal_tb.pdf
, The MathWorks Inc., pp. 271 (683).
Mayne, D.Q., Rawlings, J.B., Rao, C.V., Scokaert, P.O.M. (2000), Constrained model
predictive control: Stability and optimality, Automatica, 36, pp. 789 – 814.
185/188
Mills, A.L. (1983), Computation and modeling techniques In: Handbook of solvent
extraction, (Eds. Lo, T.C., Baird, M.H.I., Hanson, C.), Wiley & Sons, New York,
pp.841 – 852.
Mooiman, M.B, Sole, K.C., Kinneberg, D.J. (2005), Challenging the traditional
hydrometallurgy curriculum – an industry perspective, Hydrometallurgy, 79, pp. 80 –
88.
Morari, M., Lee, J.H. (1999), Model predictive control: past, present and future,
Computers and Chemical Engineering, 23, pp. 667 – 682.
Munnik, E, Singh, H., Uys, T., Bellino, M., du Plessis, J., Fraser, K. and Harris, G.B.
(2003), Development and implementation of a novel pressure leach process for the
recovery of cobalt and copper at Chambishi Zambia, The Journal of The South
African Institute of Mining and Metallurgy, 103, pp. 1-10.
Nunez, E., Desbiens, A., del Villar, R. and Duchesne, C. (2006), Multivariable
predictive control of a pilot flotation column: Part2: Identification and control, In:
Proceedings of Mineral process modeling, simulation and control, pp. 291 – 301.
Nyman, B., Ekman, E., Kuusisto, R. and Pekkala, P., (2003) The OutoCompact SX
Approach to Copper Solvent Extraction, JOM, pp. 27 – 30.
Ogata, K. (1987), Discrete-time control systems, Prentice Hall Inc., New Jersey, pp.
625-646 (984).
186/188
Orfanidis, S.J. (1990), Optimum signal processing, 2nd ed., McGraw Hill, New York,
pp. 30 – 33 (590).
Pinto, G.A., Durao, F.O., Fiuza, A.M.A., Guimaraes, M.M.B.L., Madureira, C.M.N.
(2004), Design optimisation study of solvent extraction: chemical reaction,mass
transfer and mixer–settler hydrodynamics, Hydrometallurgy, 74, pp.131–147.
Qin S.J., Badgwell, T.A. (2003), A survey of industrial model predictive control
technology, Control engineering practice, 11, pp. 733-764.
Rawlings, J.B. (2000), Tutorial overview of model predictive control, IEEE Control
Systems Magazine, 20, pp.38-52.
Ray, W.H. (1989), Advanced Process Control, Butterworth Publishers, Boston, pp.
28-30 (376).
Ray, H.S., Sridhar, R., Abraham, K.P (1985), Extraction of Nonferrous Metals,
Affiliated East-West Press Private Ltd, New Delhi, pp. 531.
Ritcey, G.M. (2002), Some design and operating problems encountered in solvent
extraction plants, In: Proceedings of the International Solvent Extraction Conference,
ISEC 2002 (eds. Sole, K.C., Cole, P.M., Preston, J.S. and Robinson, D.J.),
Johannesburg, South Africal Instritute of Mining and Metallurgy, pp. 871- 878.
Rydberg, J., Musikas, C., Choppin, G.R., Solvent extraction principles and practice,
M. Dekker, New York,2004, 584 p.
Salem, A.B. and Sheirah, M.A. (1990), Dynamic Behavior of mixer-settlers, The
Canadian Journal of Chemical Engineering, 68, pp. 867 – 875.
Sole, K.C., Feather, A.M., Cole, P.M. (2005), Solvent extraction in southern Africa:
An update of some recent hydrometallurgical developments, Hydrometallurgy, 78, pp.
52-78.
Suontaka, V.; Rantala, A.; Jämsä-Jounela, S.-L., (2004),A Process Monitoring System
for Copper Solvent Extraction, In: Proceedings of the 11th IFAC Symposium on
automation in mining, mineral and metal processing, (ed. Sauter, D.), Nancy, IFAC,
pp. .
Suontaka, V., Saloheimo, K., Marttila, T., Jämsä-Jounela, S.-L. (2003), A monitoring
system for copper solvent extraction – electrowinning process, In: Proceedings of
IFAC Workshop on New Technologies for Automation of Metallurgical Industry (ed.
Wei Wang), Shanghai, IFAC, pp. 305-310.
Tetlow, Peter (2003), Crud Formation and Techniques for Crud Control, Acorga
Notes, 9, pp. 5 – 7.]
Virnig, M.J. and Olafson, S.M. (2002), Improved Iron Scrubbing with a Wash Stage,
In: Proceedings of ALTA Copper-7 Conference, Perth, ALTA Metallurgical Services,
(CD).
Whyte, R.M., Schoeman, N. and Bowes, K.G. (2001), Processing of Konkola copper
concentrates and Chingola refractory ore in a fully integrated hydrometallurgical pilot
plant circuit, The Journal of The South African Institute of Mining and Metallurgy,
101, pp. 427-436.
Wilkinson, W.L., Ingham, J. (1983), Dynamic Behaviour and Control, In: Handbook
of Solvent Extraction, (Eds. Lo, T.C., Baird, M.H.I., Hanson, C.), Wiley & Sons, New
York, pp.853 – 886.
ISBN 978-951-22-9055-0
ISBN 978-951-22-9056-7 (PDF)
ISSN 1795-2239
ISSN 1795-4584 (PDF)